Research ArticleCausation Analysis of Hazardous Material Road TransportationAccidents by Bayesian Network Using Genie
Xiaoli Ma Yingying Xing and Jian Lu
The Key Laboratory of Road and Traffic Engineering Ministry of Education Tongji University 4800 Caorsquoan RoadShanghai 201804 China
Correspondence should be addressed to Jian Lu jianjohnlutongjieducn
Received 1 April 2018 Accepted 28 May 2018 Published 5 August 2018
Academic Editor Xiaobo Qu
Copyright copy 2018 XiaoliMa et alThis is an open access article distributed under theCreative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
With the increase of hazardous materials (Hazmat) demand and transportation frequent Hazmat road transportation accidentshad arisen the widespread concern in the community Thus it is necessary to analyze the risk factorsrsquo implications which wouldmake the safety of Hazmat transportation evolve from ldquopassive typerdquo to ldquoactive typerdquo In order to explore the influence of riskfactors resulting in accidents and predict the occurrence of accidents under the combination of risk factors 839 accidents thathave occurred for the period 2015ndash2016 were collected and examined The Bayesian network structure was established by expertsrsquoknowledge using Dempster-Shafer evidence theory Parameter learning was conducted by the Expectation-Maximization (EM)algorithm in Genie 20 The two main results could be likely to obtain the following (1) The Bayesian network model can explorethe most probable factor or combination leading to the accident which calculated the posterior probability of each risk factorFor example the importance of three or more vehicles in an accident leading to the severe accident is higher than less vehiclesand in the absence of other evidences the most probable reasons for ldquoexplosion accidentrdquo are vehicles carrying flammable liquidslarger quantity Hazmat vehicle failure and transporting in autumn (2) The model can predict the occurrence of accident bysetting the influence degrees of specific factor Such that the probability of rear-end accidents caused by ldquospeedingrdquo is 042 and theprobability could reach up to 097 when the driver is speeding at the low-class roads Moreover the complex logical relationship inHazmat road transportation accidents could be obtained and the uncertain relation among various risk factors could be expressedThese findings could provide theoretical support for transportation corporations and government department on taking effectivemeasures to reduce the risk of Hazmat road transportation
1 Introduction
In recent years the demand for hazardous materials (Haz-mat) has increased resulting in increasing transportationrequirement More than 95 Hazmat require off-site trans-portation inChina and 63 are transported by road in Brazilas well as 90 in the United States [1 2] However Hazmatcould provide the great convenience for peoplersquos life but alsosignificant risks to environment and human health exist Forinstance a total of 3744 heavy trucks were involved in severeaccidents in the United States of which 3 were carryingHazmat [3] On January 11 2015 a tanker truck carryinggasoline collided with a bus in Pakistan causing 57 deathsAnd a tanker truck carrying liquid ammonia collided with avan resulting in a large-scale spill of liquid ammonia leadingto 28 deaths and the number of poisons was up to 350 on
March 29 2005 which arose on negative social impacts inChina [4]
Hazmat transportation accidents would be able to pro-duce catastrophic influence on human health public safetyenvironment and property due to the special characteristicof Hazmat attracting more attention from general public andgovernment on the management of Hazmat road transporta-tion Thus how to improve the transportation condition andreduce the risk of transportation have become important andurgent problems for the industrial development A growingamount studies about Hazmat transportation and productionhave been conducted [5ndash7] Therefore the need for investi-gating risk factors that contribute to Hazmat accident and therelationship of risk factors are highlighted to reduce the riskof Hazmat transportation To that end the effective methodto describe and evaluate the accident process is causation
HindawiJournal of Advanced TransportationVolume 2018 Article ID 6248105 12 pageshttpsdoiorg10115520186248105
2 Journal of Advanced Transportation
analysis which could be used to determine government pri-orities related to the implementation of prevention measures[8] And causation analysis also could provide the theoreticalsupport for actionable information of controlling over therisk factors for the transportation corporations In additionexploring the most probable factor or combination leading toaccidents and predicting accidents are the important researchtopics in the field of Hazmat safety reducing the frequencyand severity of accidents
2 Literature Review
The purpose of this study is to explore risk factors to reducethe risk of Hazmat road transportation Many studies havebeen conducted by using statistical methods Haastrup andBrockhoff [9] statistically analyzed the cases of Hazmataccidents in Western Europe and 39 of accidents occurredduring transportation in 682 accidents the consequenceincluded fatality A study about Hazmat transportation acci-dents divided risk factors into human vehicle packingtransportation facilities road conditions and environmentalconditions [5] Shen [10] studied 708 accidents with Hazmatin China from 2004 to 2011 and found that accidents easilyoccurred at expressways and the higher probability of spillaccident is associated with accident type Fang et al [11]concluded that speeding was the main reason for Hazmattransportation accidents through the analysis of accident databetween 1999 and 2013
Although statistical methods could analyze the relation-ships between accidents and the risk factors they cannotaccount for the interplay among different factors and fail toreflect the fact that an accident is not usually the result ofa single factor [12] The use of causation analysis theory foraccidents could extract the accident mechanism and accidentmodels froma large number of typical accidents For instanceJason et al [13] conducted the study about the influence ofvehicle occupant driver and environmental characteristicson accident injuries involved with heavy-duty trucks andthe conclusion was obtained by using the heteroskedasticordered probit models which showed that the likelihood ofsevere accident is estimated to rise with the more vehiclesinvolved in accident Uddin and Huynh [14] used an orderedprobit model to explore the relationship among driversvehicles roadways environment temporal characteristicsand the severity of accident There was a study by using logitmodel to study the driverrsquos behaviors effect on accidents andthe results indicated that the more significant risk factorswere speeding not using seatbelt driversrsquo age and driverswith no valid license [15] In addition the Bayesian networkand tree-based methods were considered to explore deeperaccident mechanisms which is increasingly utilized in trafficaccidents analysis For instance Ona et al [16] classifiedtraffic accidents based on the severity of injuries by usingthe Bayesian networks the factors associated with fatal orsevere accidents were identified by inference such as accidenttype the driverrsquos age and lighting In order to simplify themodelMujalli et al [17] usedBayesiannetworks to reduce thenumber of variables in the study of analyzing the accidentsseverity on rural roads and the result showed that the number
of variables could reduce up to 60 (the variables consideredare accident type age atmospheric factors gender lightingnumber of injured and occupants involved) maintainingthe good performance of models Zhao et al [18] pointedthat the three most significant factors influencing Hazmattransportation by applying Bayesian networks were humanfactors the transport vehicles and facilities and the pack-aging and loading of Hazmat Chen et al [19] analyzed thebetween-accident variance and within-accident correlationsby using Bayesian network and explored the risk factorsinfluencing accidents and their heterogeneous impacts onaccident severity in rural roads And in order to improvethe efficiency of emergency rescue of Hazmat transportationroad accidents a study was conducted to evaluate the timeof accidents dealing based on the Bayesian network [20] Inaddition the Bayesian network model could also be used todescribe the probability and risk of accidents [21ndash24]
However despite many studies on the traffic accidentsand Hazmat accidents most of them are studied based onthe analysis of specific isolated and single factor [25 26]Moreover the characteristic of Hazmat was not taken intoconsideration during the analysis of accidents limiting thestudies of risk factors in Hazmat road transportation Inaddition the statistical methods could reveal the inherentrules on the occurrence of accidents but the relationshipof risk factors was not observed which cannot reflectthe accident mechanism The application of causal anal-ysis model (such as Bayesian network) can explain thecorrelation between risk factors and further explain theaccident mechanism but the Bayesian network structuremay exist subjectivity due to the expertsrsquo knowledge leadingto incorrect description of relationships between nodes inthe Bayesian network structure Therefore Hazmat roadtransportation accidents in China from 2015 to 2016 areconsidered as the research object to explore the potentialrisk factors of accidents based on expertsrsquo knowledge TheBayesian network is used to explore the most probablefactor or combination leading to accident and determine thecorrelation between the risk factors providing the decision-making basis for Hazmat transportation corporations andgovernment departments to reduce the risk of Hazmattransportation
3 Database
TheHazmat transportation accident data was obtained fromState Work Accident Briefing System and Chemical Acci-dents Information Network for two years (2015-2016) inChina and the weather data was obtained from the ChinaMeteorological Administration The regional distribution ofHazmat transportation accidents is shown in Figure 1 Thedatabase considered in the study contains 839 records andeach record contains detailed information including the datetime location type of accidents type and number of vehi-cles involved in accident driver characteristic the quantityand categories of Hazmat accident consequence causes ofaccident and a detailed description of the accident Sixteenvariables extracted from the database were considered as thesignificant factors which are shown in Table 1
Accident information is accident type (rear-end side-swipe rollover collision and vehicle failure) and accidentconsequence (explosion fire spill and nonspill) Previousstudies [14 27] divided injury severity into five categoriesthe accident severity in the paper is considered as no injurysevere injury and fatality Simplified classification of accidentseverity could ease the issue of potential relationship ofrelated consequences of an accident and ensure the sufficientsample size for the Bayesian network model [28 29] In thepaper the simplified classification of accident severity wouldobtain the better results
Hazmat information is Hazmat categories and quantity ofHazmat transportation
Driver information is characteristics of the driver such asage and behavior
Location information is road surface condition and acci-dent location (such as Group one Group two Group threeand Group four) the special road section including intersec-tion freeway service areas toll stations and gas stations areconsidered in the study
Vehicle information is type and number of vehiclesinvolved in accident
Environment information is time distribution of accident(hour day andmonth) visibility (dawn 500 to 659 am day700 am to 459 pm dusk 500 to 659 pm and dark 700 pmto 459 am) and weather conditions (sunny cloudy rainy andsnowy and fog and haze) [18]
4 Methodology
41 Definition of Bayesian Network Bayesian network isconsidered as the effective method to describe the causalitybetween the risk factors and the output in the system alsoreferred to as the belief network The Bayesian network is
a Directed Acyclic Graph (DAG) and nodes represent vari-able status while the directed edges represent dependenciesbetween variables The relationship or confidence coefficientbetween variables could be described by using ConditionalProbability Table (CPT) The Bayesian formula is consideredas the basis for the Bayesian network model which could beexpressed as
where 119875(119883 | 119884) is the probability of 119883 under the conditionof a known event 119884 119875(119884 | 119883) is the conditional probabilityof 119884 at the occurrence of119883 And the joint distribution of tworandom variables 119883 and 119884 can be expressed as
where 119875(119883) is called the prior probability and 119875(119884 | 119883)is the posterior probability Combined with the chain rulesreducing the complexity of the probability model the jointdistribution of n variables is
where119883 = 1198831 1198832 sdot sdot sdot 119883119899 setting 119878 is a network structureP is a set of local probability distributions associated witheach variable 119883119894 denotes the variable node and 119875119886119903119890119899119905(119883119894)denotes the father node of 119883119894 in 119878
4 Journal of Advanced Transportation
Table 1 Variables of Hazmat road transportation accidents
Factors Variables Variables description Discretization Frequency Percentage
The construction of the Bayesian network model consistsof following steps
(1) Parameter determination analyze the risk factors ofHazmat road transportation and determine the variablesneeded for modeling (nodes of the Bayesian network) whichcould be shown in Table 1
(2) Structure learning determine the dependencies orindependencies relationships between variables (nodes) sothat a directed acyclic network structure was constructed
(3) Parameter learning based on the given Bayesiannetwork structure determine the CPT for each node and thedependence relationship between random variables could bedescribed quantitatively
42 Structure Learning The scientific network structureneeds continuous iterations At present there are three meth-ods to construct a Bayesian network structure [30] (1) Con-struct the network structure subjectively through expertsrsquoknowledge (2) Determine the network structure objectivelyvia the analysis of data (3) Construct the network structurebased on expertsrsquo knowledge and data analysis The methodused in the paper for accident causation analysis is thatestablishing a preliminary Bayesian network structure basedon the model assumption and then the network structure isadjusted with expertsrsquo knowledge and data analysis avoidingthe disadvantage of strong subjectivity and enormous amountof data computing The Bayesian network structure is con-structed as shown in Figure 2
Steps for Building a Bayesian Network Structure(1) Establish a preliminary Bayesian network structure
based on the assumptions of model(2) Use Delphi method to determine the relationship
between risk factors In general there are four possiblerelationships between variables
(A) 119865119894 directly lead to 119865119895 which could be represented as119865119894 997888rarr 119865119895(B) 119865119895 directly lead to 119865119894 which could be represented as119865119894 larr997888 119865119895
(C) The relationship between variables cannot be deter-mined which could be represented as 119865119894 larrrarr 119865119895
(D) There is no relationship between variables whichcould be represented as 119865119894 | 119865119895
(3) Synthesize results from multiple experts D-S evi-dence theory is used to reduce the subjectivity of expertsrsquoknowledge and the correlation between variables could bedetermined The Dempster synthesis rule formula could beexpressed as
where A represents the possible relationship between vari-ables119898119894 represents the mass function equaling to the expertopinions and 119899 represents the number of experts
(4) As the relationship of variables cannot be obtainedby Delphi and D-S evidence theory the mutual informationvalue of variables should be calculated And the entropy canbe expressed as
Conditional entropy is a measure of the uncertainty of arandom variable 119865119894 under the condition of giving 119865119895 whichcan be expressed as
6 Journal of Advanced Transportation
driver_agedriver_behavior
weatherroad_surface
visibility
accident_location
accident_typespecial_section
weekly_distribution
total_vehicles_involved_in_accident
quantity_of_hazmat
hazamat_categories
season accident_consequence
severity_of_accident
type_of_vehicle
Figure 2 The Bayesian network structure for Hazmat road transportation accidents
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
2 Journal of Advanced Transportation
analysis which could be used to determine government pri-orities related to the implementation of prevention measures[8] And causation analysis also could provide the theoreticalsupport for actionable information of controlling over therisk factors for the transportation corporations In additionexploring the most probable factor or combination leading toaccidents and predicting accidents are the important researchtopics in the field of Hazmat safety reducing the frequencyand severity of accidents
2 Literature Review
The purpose of this study is to explore risk factors to reducethe risk of Hazmat road transportation Many studies havebeen conducted by using statistical methods Haastrup andBrockhoff [9] statistically analyzed the cases of Hazmataccidents in Western Europe and 39 of accidents occurredduring transportation in 682 accidents the consequenceincluded fatality A study about Hazmat transportation acci-dents divided risk factors into human vehicle packingtransportation facilities road conditions and environmentalconditions [5] Shen [10] studied 708 accidents with Hazmatin China from 2004 to 2011 and found that accidents easilyoccurred at expressways and the higher probability of spillaccident is associated with accident type Fang et al [11]concluded that speeding was the main reason for Hazmattransportation accidents through the analysis of accident databetween 1999 and 2013
Although statistical methods could analyze the relation-ships between accidents and the risk factors they cannotaccount for the interplay among different factors and fail toreflect the fact that an accident is not usually the result ofa single factor [12] The use of causation analysis theory foraccidents could extract the accident mechanism and accidentmodels froma large number of typical accidents For instanceJason et al [13] conducted the study about the influence ofvehicle occupant driver and environmental characteristicson accident injuries involved with heavy-duty trucks andthe conclusion was obtained by using the heteroskedasticordered probit models which showed that the likelihood ofsevere accident is estimated to rise with the more vehiclesinvolved in accident Uddin and Huynh [14] used an orderedprobit model to explore the relationship among driversvehicles roadways environment temporal characteristicsand the severity of accident There was a study by using logitmodel to study the driverrsquos behaviors effect on accidents andthe results indicated that the more significant risk factorswere speeding not using seatbelt driversrsquo age and driverswith no valid license [15] In addition the Bayesian networkand tree-based methods were considered to explore deeperaccident mechanisms which is increasingly utilized in trafficaccidents analysis For instance Ona et al [16] classifiedtraffic accidents based on the severity of injuries by usingthe Bayesian networks the factors associated with fatal orsevere accidents were identified by inference such as accidenttype the driverrsquos age and lighting In order to simplify themodelMujalli et al [17] usedBayesiannetworks to reduce thenumber of variables in the study of analyzing the accidentsseverity on rural roads and the result showed that the number
of variables could reduce up to 60 (the variables consideredare accident type age atmospheric factors gender lightingnumber of injured and occupants involved) maintainingthe good performance of models Zhao et al [18] pointedthat the three most significant factors influencing Hazmattransportation by applying Bayesian networks were humanfactors the transport vehicles and facilities and the pack-aging and loading of Hazmat Chen et al [19] analyzed thebetween-accident variance and within-accident correlationsby using Bayesian network and explored the risk factorsinfluencing accidents and their heterogeneous impacts onaccident severity in rural roads And in order to improvethe efficiency of emergency rescue of Hazmat transportationroad accidents a study was conducted to evaluate the timeof accidents dealing based on the Bayesian network [20] Inaddition the Bayesian network model could also be used todescribe the probability and risk of accidents [21ndash24]
However despite many studies on the traffic accidentsand Hazmat accidents most of them are studied based onthe analysis of specific isolated and single factor [25 26]Moreover the characteristic of Hazmat was not taken intoconsideration during the analysis of accidents limiting thestudies of risk factors in Hazmat road transportation Inaddition the statistical methods could reveal the inherentrules on the occurrence of accidents but the relationshipof risk factors was not observed which cannot reflectthe accident mechanism The application of causal anal-ysis model (such as Bayesian network) can explain thecorrelation between risk factors and further explain theaccident mechanism but the Bayesian network structuremay exist subjectivity due to the expertsrsquo knowledge leadingto incorrect description of relationships between nodes inthe Bayesian network structure Therefore Hazmat roadtransportation accidents in China from 2015 to 2016 areconsidered as the research object to explore the potentialrisk factors of accidents based on expertsrsquo knowledge TheBayesian network is used to explore the most probablefactor or combination leading to accident and determine thecorrelation between the risk factors providing the decision-making basis for Hazmat transportation corporations andgovernment departments to reduce the risk of Hazmattransportation
3 Database
TheHazmat transportation accident data was obtained fromState Work Accident Briefing System and Chemical Acci-dents Information Network for two years (2015-2016) inChina and the weather data was obtained from the ChinaMeteorological Administration The regional distribution ofHazmat transportation accidents is shown in Figure 1 Thedatabase considered in the study contains 839 records andeach record contains detailed information including the datetime location type of accidents type and number of vehi-cles involved in accident driver characteristic the quantityand categories of Hazmat accident consequence causes ofaccident and a detailed description of the accident Sixteenvariables extracted from the database were considered as thesignificant factors which are shown in Table 1
Accident information is accident type (rear-end side-swipe rollover collision and vehicle failure) and accidentconsequence (explosion fire spill and nonspill) Previousstudies [14 27] divided injury severity into five categoriesthe accident severity in the paper is considered as no injurysevere injury and fatality Simplified classification of accidentseverity could ease the issue of potential relationship ofrelated consequences of an accident and ensure the sufficientsample size for the Bayesian network model [28 29] In thepaper the simplified classification of accident severity wouldobtain the better results
Hazmat information is Hazmat categories and quantity ofHazmat transportation
Driver information is characteristics of the driver such asage and behavior
Location information is road surface condition and acci-dent location (such as Group one Group two Group threeand Group four) the special road section including intersec-tion freeway service areas toll stations and gas stations areconsidered in the study
Vehicle information is type and number of vehiclesinvolved in accident
Environment information is time distribution of accident(hour day andmonth) visibility (dawn 500 to 659 am day700 am to 459 pm dusk 500 to 659 pm and dark 700 pmto 459 am) and weather conditions (sunny cloudy rainy andsnowy and fog and haze) [18]
4 Methodology
41 Definition of Bayesian Network Bayesian network isconsidered as the effective method to describe the causalitybetween the risk factors and the output in the system alsoreferred to as the belief network The Bayesian network is
a Directed Acyclic Graph (DAG) and nodes represent vari-able status while the directed edges represent dependenciesbetween variables The relationship or confidence coefficientbetween variables could be described by using ConditionalProbability Table (CPT) The Bayesian formula is consideredas the basis for the Bayesian network model which could beexpressed as
where 119875(119883 | 119884) is the probability of 119883 under the conditionof a known event 119884 119875(119884 | 119883) is the conditional probabilityof 119884 at the occurrence of119883 And the joint distribution of tworandom variables 119883 and 119884 can be expressed as
where 119875(119883) is called the prior probability and 119875(119884 | 119883)is the posterior probability Combined with the chain rulesreducing the complexity of the probability model the jointdistribution of n variables is
where119883 = 1198831 1198832 sdot sdot sdot 119883119899 setting 119878 is a network structureP is a set of local probability distributions associated witheach variable 119883119894 denotes the variable node and 119875119886119903119890119899119905(119883119894)denotes the father node of 119883119894 in 119878
4 Journal of Advanced Transportation
Table 1 Variables of Hazmat road transportation accidents
Factors Variables Variables description Discretization Frequency Percentage
The construction of the Bayesian network model consistsof following steps
(1) Parameter determination analyze the risk factors ofHazmat road transportation and determine the variablesneeded for modeling (nodes of the Bayesian network) whichcould be shown in Table 1
(2) Structure learning determine the dependencies orindependencies relationships between variables (nodes) sothat a directed acyclic network structure was constructed
(3) Parameter learning based on the given Bayesiannetwork structure determine the CPT for each node and thedependence relationship between random variables could bedescribed quantitatively
42 Structure Learning The scientific network structureneeds continuous iterations At present there are three meth-ods to construct a Bayesian network structure [30] (1) Con-struct the network structure subjectively through expertsrsquoknowledge (2) Determine the network structure objectivelyvia the analysis of data (3) Construct the network structurebased on expertsrsquo knowledge and data analysis The methodused in the paper for accident causation analysis is thatestablishing a preliminary Bayesian network structure basedon the model assumption and then the network structure isadjusted with expertsrsquo knowledge and data analysis avoidingthe disadvantage of strong subjectivity and enormous amountof data computing The Bayesian network structure is con-structed as shown in Figure 2
Steps for Building a Bayesian Network Structure(1) Establish a preliminary Bayesian network structure
based on the assumptions of model(2) Use Delphi method to determine the relationship
between risk factors In general there are four possiblerelationships between variables
(A) 119865119894 directly lead to 119865119895 which could be represented as119865119894 997888rarr 119865119895(B) 119865119895 directly lead to 119865119894 which could be represented as119865119894 larr997888 119865119895
(C) The relationship between variables cannot be deter-mined which could be represented as 119865119894 larrrarr 119865119895
(D) There is no relationship between variables whichcould be represented as 119865119894 | 119865119895
(3) Synthesize results from multiple experts D-S evi-dence theory is used to reduce the subjectivity of expertsrsquoknowledge and the correlation between variables could bedetermined The Dempster synthesis rule formula could beexpressed as
where A represents the possible relationship between vari-ables119898119894 represents the mass function equaling to the expertopinions and 119899 represents the number of experts
(4) As the relationship of variables cannot be obtainedby Delphi and D-S evidence theory the mutual informationvalue of variables should be calculated And the entropy canbe expressed as
Conditional entropy is a measure of the uncertainty of arandom variable 119865119894 under the condition of giving 119865119895 whichcan be expressed as
6 Journal of Advanced Transportation
driver_agedriver_behavior
weatherroad_surface
visibility
accident_location
accident_typespecial_section
weekly_distribution
total_vehicles_involved_in_accident
quantity_of_hazmat
hazamat_categories
season accident_consequence
severity_of_accident
type_of_vehicle
Figure 2 The Bayesian network structure for Hazmat road transportation accidents
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Accident information is accident type (rear-end side-swipe rollover collision and vehicle failure) and accidentconsequence (explosion fire spill and nonspill) Previousstudies [14 27] divided injury severity into five categoriesthe accident severity in the paper is considered as no injurysevere injury and fatality Simplified classification of accidentseverity could ease the issue of potential relationship ofrelated consequences of an accident and ensure the sufficientsample size for the Bayesian network model [28 29] In thepaper the simplified classification of accident severity wouldobtain the better results
Hazmat information is Hazmat categories and quantity ofHazmat transportation
Driver information is characteristics of the driver such asage and behavior
Location information is road surface condition and acci-dent location (such as Group one Group two Group threeand Group four) the special road section including intersec-tion freeway service areas toll stations and gas stations areconsidered in the study
Vehicle information is type and number of vehiclesinvolved in accident
Environment information is time distribution of accident(hour day andmonth) visibility (dawn 500 to 659 am day700 am to 459 pm dusk 500 to 659 pm and dark 700 pmto 459 am) and weather conditions (sunny cloudy rainy andsnowy and fog and haze) [18]
4 Methodology
41 Definition of Bayesian Network Bayesian network isconsidered as the effective method to describe the causalitybetween the risk factors and the output in the system alsoreferred to as the belief network The Bayesian network is
a Directed Acyclic Graph (DAG) and nodes represent vari-able status while the directed edges represent dependenciesbetween variables The relationship or confidence coefficientbetween variables could be described by using ConditionalProbability Table (CPT) The Bayesian formula is consideredas the basis for the Bayesian network model which could beexpressed as
where 119875(119883 | 119884) is the probability of 119883 under the conditionof a known event 119884 119875(119884 | 119883) is the conditional probabilityof 119884 at the occurrence of119883 And the joint distribution of tworandom variables 119883 and 119884 can be expressed as
where 119875(119883) is called the prior probability and 119875(119884 | 119883)is the posterior probability Combined with the chain rulesreducing the complexity of the probability model the jointdistribution of n variables is
where119883 = 1198831 1198832 sdot sdot sdot 119883119899 setting 119878 is a network structureP is a set of local probability distributions associated witheach variable 119883119894 denotes the variable node and 119875119886119903119890119899119905(119883119894)denotes the father node of 119883119894 in 119878
4 Journal of Advanced Transportation
Table 1 Variables of Hazmat road transportation accidents
Factors Variables Variables description Discretization Frequency Percentage
The construction of the Bayesian network model consistsof following steps
(1) Parameter determination analyze the risk factors ofHazmat road transportation and determine the variablesneeded for modeling (nodes of the Bayesian network) whichcould be shown in Table 1
(2) Structure learning determine the dependencies orindependencies relationships between variables (nodes) sothat a directed acyclic network structure was constructed
(3) Parameter learning based on the given Bayesiannetwork structure determine the CPT for each node and thedependence relationship between random variables could bedescribed quantitatively
42 Structure Learning The scientific network structureneeds continuous iterations At present there are three meth-ods to construct a Bayesian network structure [30] (1) Con-struct the network structure subjectively through expertsrsquoknowledge (2) Determine the network structure objectivelyvia the analysis of data (3) Construct the network structurebased on expertsrsquo knowledge and data analysis The methodused in the paper for accident causation analysis is thatestablishing a preliminary Bayesian network structure basedon the model assumption and then the network structure isadjusted with expertsrsquo knowledge and data analysis avoidingthe disadvantage of strong subjectivity and enormous amountof data computing The Bayesian network structure is con-structed as shown in Figure 2
Steps for Building a Bayesian Network Structure(1) Establish a preliminary Bayesian network structure
based on the assumptions of model(2) Use Delphi method to determine the relationship
between risk factors In general there are four possiblerelationships between variables
(A) 119865119894 directly lead to 119865119895 which could be represented as119865119894 997888rarr 119865119895(B) 119865119895 directly lead to 119865119894 which could be represented as119865119894 larr997888 119865119895
(C) The relationship between variables cannot be deter-mined which could be represented as 119865119894 larrrarr 119865119895
(D) There is no relationship between variables whichcould be represented as 119865119894 | 119865119895
(3) Synthesize results from multiple experts D-S evi-dence theory is used to reduce the subjectivity of expertsrsquoknowledge and the correlation between variables could bedetermined The Dempster synthesis rule formula could beexpressed as
where A represents the possible relationship between vari-ables119898119894 represents the mass function equaling to the expertopinions and 119899 represents the number of experts
(4) As the relationship of variables cannot be obtainedby Delphi and D-S evidence theory the mutual informationvalue of variables should be calculated And the entropy canbe expressed as
Conditional entropy is a measure of the uncertainty of arandom variable 119865119894 under the condition of giving 119865119895 whichcan be expressed as
6 Journal of Advanced Transportation
driver_agedriver_behavior
weatherroad_surface
visibility
accident_location
accident_typespecial_section
weekly_distribution
total_vehicles_involved_in_accident
quantity_of_hazmat
hazamat_categories
season accident_consequence
severity_of_accident
type_of_vehicle
Figure 2 The Bayesian network structure for Hazmat road transportation accidents
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
The construction of the Bayesian network model consistsof following steps
(1) Parameter determination analyze the risk factors ofHazmat road transportation and determine the variablesneeded for modeling (nodes of the Bayesian network) whichcould be shown in Table 1
(2) Structure learning determine the dependencies orindependencies relationships between variables (nodes) sothat a directed acyclic network structure was constructed
(3) Parameter learning based on the given Bayesiannetwork structure determine the CPT for each node and thedependence relationship between random variables could bedescribed quantitatively
42 Structure Learning The scientific network structureneeds continuous iterations At present there are three meth-ods to construct a Bayesian network structure [30] (1) Con-struct the network structure subjectively through expertsrsquoknowledge (2) Determine the network structure objectivelyvia the analysis of data (3) Construct the network structurebased on expertsrsquo knowledge and data analysis The methodused in the paper for accident causation analysis is thatestablishing a preliminary Bayesian network structure basedon the model assumption and then the network structure isadjusted with expertsrsquo knowledge and data analysis avoidingthe disadvantage of strong subjectivity and enormous amountof data computing The Bayesian network structure is con-structed as shown in Figure 2
Steps for Building a Bayesian Network Structure(1) Establish a preliminary Bayesian network structure
based on the assumptions of model(2) Use Delphi method to determine the relationship
between risk factors In general there are four possiblerelationships between variables
(A) 119865119894 directly lead to 119865119895 which could be represented as119865119894 997888rarr 119865119895(B) 119865119895 directly lead to 119865119894 which could be represented as119865119894 larr997888 119865119895
(C) The relationship between variables cannot be deter-mined which could be represented as 119865119894 larrrarr 119865119895
(D) There is no relationship between variables whichcould be represented as 119865119894 | 119865119895
(3) Synthesize results from multiple experts D-S evi-dence theory is used to reduce the subjectivity of expertsrsquoknowledge and the correlation between variables could bedetermined The Dempster synthesis rule formula could beexpressed as
where A represents the possible relationship between vari-ables119898119894 represents the mass function equaling to the expertopinions and 119899 represents the number of experts
(4) As the relationship of variables cannot be obtainedby Delphi and D-S evidence theory the mutual informationvalue of variables should be calculated And the entropy canbe expressed as
Conditional entropy is a measure of the uncertainty of arandom variable 119865119894 under the condition of giving 119865119895 whichcan be expressed as
6 Journal of Advanced Transportation
driver_agedriver_behavior
weatherroad_surface
visibility
accident_location
accident_typespecial_section
weekly_distribution
total_vehicles_involved_in_accident
quantity_of_hazmat
hazamat_categories
season accident_consequence
severity_of_accident
type_of_vehicle
Figure 2 The Bayesian network structure for Hazmat road transportation accidents
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
The construction of the Bayesian network model consistsof following steps
(1) Parameter determination analyze the risk factors ofHazmat road transportation and determine the variablesneeded for modeling (nodes of the Bayesian network) whichcould be shown in Table 1
(2) Structure learning determine the dependencies orindependencies relationships between variables (nodes) sothat a directed acyclic network structure was constructed
(3) Parameter learning based on the given Bayesiannetwork structure determine the CPT for each node and thedependence relationship between random variables could bedescribed quantitatively
42 Structure Learning The scientific network structureneeds continuous iterations At present there are three meth-ods to construct a Bayesian network structure [30] (1) Con-struct the network structure subjectively through expertsrsquoknowledge (2) Determine the network structure objectivelyvia the analysis of data (3) Construct the network structurebased on expertsrsquo knowledge and data analysis The methodused in the paper for accident causation analysis is thatestablishing a preliminary Bayesian network structure basedon the model assumption and then the network structure isadjusted with expertsrsquo knowledge and data analysis avoidingthe disadvantage of strong subjectivity and enormous amountof data computing The Bayesian network structure is con-structed as shown in Figure 2
Steps for Building a Bayesian Network Structure(1) Establish a preliminary Bayesian network structure
based on the assumptions of model(2) Use Delphi method to determine the relationship
between risk factors In general there are four possiblerelationships between variables
(A) 119865119894 directly lead to 119865119895 which could be represented as119865119894 997888rarr 119865119895(B) 119865119895 directly lead to 119865119894 which could be represented as119865119894 larr997888 119865119895
(C) The relationship between variables cannot be deter-mined which could be represented as 119865119894 larrrarr 119865119895
(D) There is no relationship between variables whichcould be represented as 119865119894 | 119865119895
(3) Synthesize results from multiple experts D-S evi-dence theory is used to reduce the subjectivity of expertsrsquoknowledge and the correlation between variables could bedetermined The Dempster synthesis rule formula could beexpressed as
where A represents the possible relationship between vari-ables119898119894 represents the mass function equaling to the expertopinions and 119899 represents the number of experts
(4) As the relationship of variables cannot be obtainedby Delphi and D-S evidence theory the mutual informationvalue of variables should be calculated And the entropy canbe expressed as
Conditional entropy is a measure of the uncertainty of arandom variable 119865119894 under the condition of giving 119865119895 whichcan be expressed as
6 Journal of Advanced Transportation
driver_agedriver_behavior
weatherroad_surface
visibility
accident_location
accident_typespecial_section
weekly_distribution
total_vehicles_involved_in_accident
quantity_of_hazmat
hazamat_categories
season accident_consequence
severity_of_accident
type_of_vehicle
Figure 2 The Bayesian network structure for Hazmat road transportation accidents
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Before obtaining 119865119895 the uncertainty of 119865119894 is 119867(119865119894) andafter obtaining 119865119895 the uncertainty of 119865119894 is119867(119865119894 | 119865119895) so thatthe difference of 119867(119865119894) and 119867(119865119894 | 119865119895) is considered as themutual information which is expressed as
43 Parameter Learning There are missing data on Hazmatroad transportation accidents the Expectation- Maximiza-tion (EM) algorithm is considered as the effective methodto perform the maximum likelihood estimation for a set ofparameters 120579 from the incomplete dataset [31ndash33] The EMalgorithm starts with randomly assigning a configuration 1205790for 120579 by the system Suppose that 120579119905 is the outcome after titerations The calculation process mainly involved two stepsExpectation Step (E-Step) and Maximization Step (M-Step)
Consider that 119863119898 is missing sample and 119883119898 is the set ofall variables with missing value in the sample 119863119898 Set 119883119898 =119909119898 and the complete dataset would be obtained by adding
119909119898 to 119863119898 All of the possible result would be considered byEM algorithm due to that 119883119898 may have more possibility sothe weight 119908119909119898 is assigned for each possible result by EMalgorithm and the weighted sample could be given by
where 119908119909119898 = 119875(119883119898 = 119909119898 | 119863119898 120579119905) and the weight rangesfrom 0 to 1E-Step suppose the log-likelihood function of 120579 based on119863119905
where 119863 = (1198631 1198632 119863119898) and 119898(120579 | 119863 120579119905) = 119898(120579 | 119863119905)is referred to as the expected log-likelihood function In theiteration due to the characteristic of 119863 which is invariantthe formula could be expressed as
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
where119898119905119894119895119896 is the sum of sample weights in the dataset 1198631199055 Results
Theguidance for the variable selection and classification werefollowed by the analysis of accident data and previous studies[6 34ndash36] In the paper sixteen variables are consideredas the significant risk factors as shown in Table 1 Thereare numerous types of software to establish the Bayesiannetwork efficiently such as Netica Genie Bayes Net Toolboxand Analytica In the paper Genie20 (developed by theDecision Systems Laboratory the University of Pittsburgh)was considered as the effective tool to finish the Bayesiannetwork parameter learning by using EM algorithm whichwould make the construction analysis and visualizationof Bayesian network be performed efficiently simplifying
the calculation And the network parameters are repeatedlyiterated by using the accident data the conditions for thetermination of calculation are as follows (1) the variation ofthe posterior probability for single risk factor is less than 1(2) the cumulative variation of posterior probability for theentire network is less than 15 The results were shown inFigure 3
51 Causal Inference The Bayesian network could be usedto calculate the posterior probability of risk factors underconditions of an accident and obtain the most likely factorsor combinations that caused accidents Set the ldquoexplosionrdquoin ldquoaccident consequencerdquo as the example to explore thecausal inference and the evidence variable is ldquoexplosionrdquoAs shown in Figure 4 the probabilities of risk factors areobtained through the update function of the Genie Andthe probability of ldquoautumnrdquo in ldquoseasonrdquo increases from22 to 35 ldquovehicle failurerdquo (referred as the tire blowoutspontaneous combustion tanker damage) in ldquoaccident typerdquoincreases from 17 to 37 the quantity of Hazmat increasesfrom 8 to 20 for the category of more than 40 tonsldquoflammable liquidsrdquo in ldquoHazmat categoriesrdquo increase from51 to 65 and the explosives increase from 3 to 8Thesefindings mean that in the absence of other evidences themost probable reasons for ldquoexplosionrdquo are vehicles carrying
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Figure 4 Posterior probability when the variable is ldquoexplosionrdquo
flammable liquids larger quantity of Hazmat vehicle failureand transporting in autumn
In addition if the ldquofatalityrdquo in the ldquoseverity of accidentrdquo isconsidered as the evidence variable the probability changeof ldquototal vehicle involved accidentrdquo could be obtained Theprobability of ldquothreerdquo increases from 4 to 11 and ldquomorethan threerdquo is increasing from 3 to 9 This may beexplained by the fact that the importance of 3 ormore vehiclesin an accident leads to the severe accident being higher thanless vehicles Moreover as for the accident consequence theprobability of ldquospillrdquo decreases meanwhile the ldquoexplosionrdquo(3 to 6) and fire (11 to 18) have increased Due to thespecial characteristic of Hazmat explosion and fire wouldcause a larger area affected and can easily result in casualtiesespecially in the urban road and higher population densities[26]
52 Accident Prediction Based on the bidirectional reason-ing not only could the Bayesian network model obtain therisk factors or the combination caused accidents but alsothe probability of accidents could be calculated under therisk factors or combination for example in Genie settingthe ldquospeedingrdquo in ldquodriver behaviorrdquo as an evidence variablemeaning that the status of evidence variable is consideredas 100 As can be seen from Figure 5 the probability of
ldquorear-endrdquo in ldquoaccident typerdquo is found to increase from 27to 42 indicating that the driversrsquo speeding could be moreprone to lead to rear-end accidentsThis is because the vehicleis difficult to control under the condition of speeding andthe braking time is longer And previous studies have shownthat driving behavior could significantly affect the severity oftraffic accidents [37ndash39]
As shown in Figure 6 in addition to ldquospeedingrdquo itis assumed that the transportation route is on low-classroads that is ldquoGroup fourrdquo in the ldquoaccident locationrdquo isconsidered as the evidence variable and the probability ofthe entire network is automatically updated It can be foundthat the probability of ldquorolloverrdquo in ldquoaccident typerdquo furtherincreases from 42 to 97 This finding shows that ldquodriverbehaviorrdquo and ldquoaccident locationrdquo would affect the probabilityof ldquorolloverrdquo accident on different degrees Therefore whenthe driver is speeding on low-class roads the more attentionshould be paid on the rollover accident
6 Discussion and Conclusions
61 Hazmat Factors Flammable liquids have the highest pos-terior probability (051) and would easily result in explosionThis could be explained by that increasing demand for theflammable liquid and decreasing reliability of transporting
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Figure 5 Accident prediction when the evidence variable is ldquospeedingrdquo
flammable liquids due to the single-mode packaging Thequantity of Hazmat transported would significantly affectthe severity of accident The larger the quantity of Hazmattransportation the larger the inertia of the transportationvehicles making it not easy to control the emergency[40] Moreover the larger quantity of Hazmat transporta-tion is prone to the serious consequences such as explo-sion and spill threatening peoplersquos health and environment[10]
62 Driver Factors Previous studies have shown the rela-tionship between driverrsquos age and the severity of accidents[27 41 42] According to the model results the youngerdriver (less than 35) would be more prone to inappropriatedriving behavior which indicates the need for carrying outeducation programs and training for younger drivers Tavriset al [43] also found that younger drivers were much morelikely to be involved in severe and fatal accidents As for thedriving behavior speeding is more likely to lead to rolloveraccident especially on the low-class road This could ascribethe small amount of lanes and the road condition defects onlow-class roads and the speeding would make Hazmat sloshor move around inside the tank which can constantly shiftthe vehicle weight leading to vehicle to rollover due to the offbalance [44 45]
63 Location Factors The model results show that ldquoGrouponerdquo (the posterior probability is 043) and ldquoGroup twordquo(the posterior probability is 040) in ldquoaccident locationrdquoare likely to be associated with severe accidents whichcould be attributed by the combination of higher averagespeed and larger speed dispersionMore importantly ldquoGrouponerdquo and ldquoGroup twordquo roads are considered as the majortransport corridors for Hazmat [10 46] In addition somespecial sections would also considered as the significant riskfactors this could be explained by the fact that there aremore interference factors (such as line of sight pedestriansand signal lights) at intersections and the greater potentialexplosion risk around the gas stations [47]
64 Environment Factors Hazmat road transportation acci-dents would easily occur at summer (the posterior prob-ability is 031) which is attributed to the characteristic ofHazmat such as flammable and explosive And the poste-rior probability of accidents occurring at weekdays is 076which could be explained by that freeway could be toll-freeon important holidays resulting in significant increase oftraffic volume which could decrease the speed of vehiclesMoreover Hazmat transportation vehicles were not allowedto drive on freeway (Pan 2013) Weather is a significantfactor for the Hazmat transportation with cloudy having the
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Figure 6 Accident prediction when the evidence variable are ldquospeedingrdquo and ldquoGroup fourrdquo
highest posterior probability (041) followed by rainy (032)This could be ascribed that the driverrsquos mood and visualwould be decreased in cloudy and rainy and the rainy wouldlower the friction coefficient of roads due to the thin filmof water existing between the road surface and tires whichcould make the road slippery increasing the braking distanceeffectively [48 49] Regarding visibility daytime has thehighest posterior probability (049) and the dark is 033Thisis because most transportation corporations are more likelyto transport Hazmat at daytime in China [50] In additionpoor visibility at night would make drivers tired resulting indriver fatigue especially from 1100 pm to 300 am [51] Inthe sample of accident data drivers are more prone to fatiguestatus accounting for 62 of total accidents from 700 pm to459 am
65 Vehicle Factors As for the total vehicles involved inaccident ldquomore than threerdquo would easily result in higherseverity of accidents And the private car involved in acci-dent would cause the severe accident Two reasons couldexplain these findings one is that more vehicles would causemore people involved in accidents resulting in more peopleinjured another one is the disparity in mass and speed oftrucks compared to other vehicles In case of an accident
lighter vehicles (such as private cars) usually absorb thegreatest part of the kinetic energy and suffer frommore severeinjury
66 Accident Factors Many studies have shown the signif-icant relationship of accidents type and severity indicatingthat the rollover accident is associatedwith the higher severityof accident [16 44] The Bayesian network results show thatrollover accident has the highest posterior probability (041)The reason could be that Hazmat sloshing or moving aroundinside the tank can constantly shift the vehicleweightmakingthe vehicle off balance causing the transportation vehicleto roll over especially during abrupt evasive maneuvers orturning the vehicle [10] In addition as for the consequenceof accident the posterior probability of spill could reachup to 081 threatening human health and environment Theresult could be explained by that Hazmat releasing couldimmediately result in poisoning and suffocation which isdifficult for people on-site to escape quickly resulting insevere and fatality accident [50]
In summary the occurrence of Hazmat road transporta-tion accidents is unexpected random dangerous and poten-tial Frequent accidents imply that it is necessary to explorerisk factors by using accident mechanism Bayesian network
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
Journal of Advanced Transportation 11
is the effective method to deal with uncertainties whichexhibit the potential hierarchical relation by the DirectedAcyclic Graph In the paper the Bayesian network wasdeveloped based on expertsrsquo knowledge and modified basedon the Hazmat road transportation accident data (N=839)in China The Bayesian network structure was established byusing Genie 20 and the results of network structure modelreveal the influence of risk factors resulting in accidents andthe relationship among risk factors The study shows that theposterior probability of the Bayesian network could provideeffective method for finding the important factors and thefactors combination of accidents These findings could pro-vide theoretical guidance which could help transportationcorporations and government departments take necessarymeasures to reduce the frequency of Hazmat accidents Moreimportantly it must be noted that the aforementioned resultswere obtained by analyzing the data sample collected fromState Work Accident Briefing System and Hazardous Chem-ical Accidents Communications which could be existinglimitations As for the further studies the conclusions shouldbe more generalizable if the dataset had larger size of sampleand accidents from multiple states
Data Availability
The data used to support the findings of this study areavailable from the corresponding author upon request
Conflicts of Interest
The authors declare that they have no conflicts of interest
Acknowledgments
This study has been supported by projects of the NationalNatural Science Foundation of China (no 71671127)
References
[1] K G Zografos and K N Androutsopoulos ldquoA decision sup-port system for integrated hazardous materials routing andemergency response decisionsrdquo Transportation Research Part CEmerging Technologies vol 16 no 6 pp 684ndash703 2008
[2] F G Cordeiro B S Bezerra A S P Peixoto andRA R RamosldquoMethodological aspects for modeling the environmental riskof transporting hazardous materials by roadrdquo TransportationResearch Part D Transport and Environment vol 44 pp 105ndash121 2016
[3] Federal Motor Carrier Safety Administration (FMCSA) ldquoLargeTruck and Bus Crash Facts 2014rdquo httpswwwfmcsadotgovsitesfmcsadotgovfilesdocsLarge-Truck-and-Bus-Crash-Facts-201428April20201629pdf 2016
[4] J Zhao L ldquoRisk Analysis of Dangerous Chemicals Transporta-tionrdquo Systems Engineering-Theory Practice vol 27 no 12 pp117ndash122 2007
[5] W Y Hua and A Tong P ldquoRisk Analysis on Road TransportSystem of Dangerous Chemicalsrdquo China Safety Science Journalvol 15 no 2 pp 8ndash12 2005
[6] L Zhao J P Wu and K Xu ldquoStatistic analysis and countermea-sures on dangerous chemical accidents in Chinardquo China SafetyScience Journal 2009
[7] J Yang F Li J Zhou L Zhang L Huang and J Bi ldquoA survey onhazardous materials accidents during road transport in Chinafrom2000 to 2008rdquo Journal of HazardousMaterials vol 184 no1-3 pp 647ndash653 2010
[8] T Kauppinen and J Rantanen ldquoWork and Health CountryProfiles and National Surveillance Indicators in OccupationalHealth and Safetyrdquo Applied Occupational amp EnvironmentalHygiene vol 17 no 9 p 603 2002
[9] P Haastrup and L Brockhoff ldquoSeverity of accidents withhazardousmaterials A comparison between transportation andfixed installationsrdquo Journal of Loss Prevention in the ProcessIndustries vol 3 no 4 pp 395ndash405 1990
[10] X Shen Y Yan X Li C Xie and L Wang ldquoAnalysis onTank Truck Accidents Involved in Road Hazardous MaterialsTransportation in Chinardquo Traffic Injury Prevention vol 15 no7 pp 762ndash768 2014
[11] K Fang G Y Ke and M Verma ldquoA routing and schedulingapproach to rail transportation of hazardous materials withdemand due datesrdquo European Journal of Operational Researchvol 261 no 1 pp 154ndash168 2017
[12] F Bird and G Germain Practical Loss Control LeadershipInternational Loss Control Institute Duluth GA USA Revisededition 1990
[13] J D LempKMKockelman andAUnnikrishnan ldquoAnalysis oflarge truck crash severity using heteroskedastic ordered probitmodelsrdquo Accident Analysis amp Prevention vol 43 no 1 pp 370ndash380 2011
[14] M Uddin and N Huynh ldquoFactors influencing injury severityof crashes involving HAZMAT trucksrdquo International Journal ofTransportation Science andTechnology vol 7 no 1 pp 1ndash9 2018
[15] E K Adanu and S Jones ldquoEffects of Human-Centered Factorson Crash Injury Severitiesrdquo Journal of Advanced Transportationvol 2017 no 1528 pp 1ndash11 2017
[16] J De Ona R O Mujalli and F J Calvo ldquoAnalysis of traffic acci-dent injury severity on Spanish rural highways using Bayesiannetworksrdquo Accident Analysis amp Prevention vol 43 no 1 pp402ndash411 2011
[17] R O Mujalli and J De Ona ldquoA method for simplifying theanalysis of traffic accidents injury severity on two-lane highwaysusing Bayesian networksrdquo Journal of Safety Research vol 42 no5 pp 317ndash326 2011
[18] L J Zhao X L Wang and Y Qian ldquoAnalysis of factors thatinfluence hazardousmaterial transportation accidents based onBayesiannetworks a case study inChinardquo Safety Science vol 50no 4 pp 1049ndash1055 2012
[19] C Chen G Zhang X C Liu et al ldquoDriver injury severityoutcome analysis in rural interstate highway crashes a two-levelBayesian logistic regression interpretationrdquoAccident Analysis ampPrevention vol 97 pp 69ndash78 2016
[20] J Chen M Zhang S Yu and J Wang ldquoA Bayesian Network forthe Transportation Accidents of Hazardous Materials HandlingTime Assessmentrdquo Procedia Engineering vol 211 pp 63ndash692018
[21] M Deublein M Schubert B T Adey J Kohler and M HFaber ldquoPrediction of road accidents a Bayesian hierarchicalapproachrdquo Accident Analysis amp Prevention vol 51 pp 274ndash2912013
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
Chemical EngineeringInternational Journal of Antennas and
Propagation
International Journal of
Hindawiwwwhindawicom Volume 2018
Hindawiwwwhindawicom Volume 2018
Navigation and Observation
International Journal of
Hindawi
wwwhindawicom Volume 2018
Advances in
Multimedia
Submit your manuscripts atwwwhindawicom
12 Journal of Advanced Transportation
[22] C Xu W Wang P Liu and Z Li ldquoCalibration of crashrisk models on freeways with limited real-time traffic datausing Bayesianmeta-analysis andBayesian inference approachrdquoAccident Analysis amp Prevention vol 85 pp 207ndash218 2015
[23] C Tang Y Yi Z Yang and J Sun ldquoRisk analysis of emergentwater pollution accidents based on a BayesianNetworkrdquo Journalof Environmental Management vol 165 pp 199ndash205 2016
[24] X Zou and W L Yue ldquoA Bayesian Network Approach toCausation Analysis of Road Accidents Using Neticardquo Journal ofAdvanced Transportation 2017
[25] R Bubbico S Di Cave B Mazzarotta and B Silvetti ldquoPrelim-inary study on the transport of hazardous materials throughtunnelsrdquoAccident Analysis amp Prevention vol 41 no 6 pp 1199ndash1205 2009
[26] R A Garrido and A C Bronfman ldquoEquity and social accept-ability in multiple hazardous materials routing through urbanareasrdquo Transportation Research Part A Policy and Practice vol102 pp 244ndash260 2016
[27] T Usman L Fu and L F Miranda-Moreno ldquoAnalysis offactors affecting winter collision severityrdquo in Meeting of theTransportation Research Board Washington DC USA 2013
[28] F Chen and S Chen ldquoInjury severities of truck drivers insingle- andmulti-vehicle accidents on rural highwaysrdquoAccidentAnalysis amp Prevention vol 43 no 5 pp 1677ndash1688 2011
[29] S Islam S L Jones and D Dye ldquoComprehensive analysisof single- and multi-vehicle large truck at-fault crashes onrural and urban roadways in Alabamardquo Accident Analysis ampPrevention vol 67 pp 148ndash158 2014
[30] Q Xiaohu L Li and Z Ying ldquoA traffic accident predictionmethod based on Bayesian network modelrdquo Computer Simula-tion vol 22 no 11 pp 230ndash232 2005
[31] S L Lauritzen ldquoThe EM algorithm for graphical associationmodels with missing datardquo Computational Statistics amp DataAnalysis vol 19 no 2 pp 191ndash201 1995
[32] F V Jensen and T D Nielsen ldquoBayesianNetworks andDecisionGraphsrdquo Technometrics vol 50 no 1 p 362 2012
[33] J Zhou W Xu X Guo and J Ding ldquoA method for modelingand analysis of directed weighted accident causation network(DWACN)rdquo Physica A Statistical Mechanics and its Applica-tions vol 437 pp 263ndash277 2015
[34] J Abellan G Lopez and J de Ona ldquoAnalysis of traffic accidentseverity using decision rules via decision treesrdquo Expert Systemswith Applications vol 40 no 15 pp 6047ndash6054 2013
[35] R O Mujalli G Lopez and L Garach ldquoBayes classifiersfor imbalanced traffic accidents datasetsrdquo Accident Analysis ampPrevention vol 88 pp 37ndash51 2016
[36] A Iranitalab and A Khattak ldquoComparison of four statisticaland machine learning methods for crash severity predictionrdquoAccident Analysis amp Prevention vol 108 pp 27ndash36 2017
[37] L Fridstroslashm and S Ingebrigtsen ldquoAn aggregate accident modelbased on pooled regional time-series datardquo Accident Analysisamp Prevention vol 23 no 5 pp 363ndash378 1991
[38] G F Ulfarsson S Kim and E T Lentz ldquoFactors affecting com-mon vehicle-to-vehicle collision types Road safety priorities inan aging societyrdquo Transportation Research Board vol 1980 no1 pp 70ndash78 2006
[39] G Zhang K K W Yau X Zhang and Y Li ldquoTraffic accidentsinvolving fatigue driving and their extent of casualtiesrdquoAccidentAnalysis amp Prevention vol 87 pp 34ndash42 2016
[40] A Vorster ldquoTransporting dangerous goods worldwide materi-als handling logisticsrdquo South African Pharmaceutical CosmeticReview 2015
[41] A P Jones and S H Joslashrgensen ldquoThe use of multilevel modelsfor the prediction of road accident outcomesrdquoAccident Analysisamp Prevention vol 35 no 1 pp 59ndash69 2003
[42] S Kaplan and C G Prato ldquoRisk factors associated with busaccident severity in the United States a generalized orderedlogit modelrdquo Journal of Safety Research vol 43 no 3 pp 171ndash180 2012
[43] D R Tavris E M Kuhn and P M Layde ldquoAge and genderpatterns in motor vehicle crash injuries Importance of type ofcrash and occupant rolerdquo Accident Analysis amp Prevention vol33 no 2 pp 167ndash172 2001
[44] C S Duncan A J Khattak and F M Council ldquoApplying theordered probit model to injury severity in truck-passenger carrear-end collisionsrdquo Transportation Research Record no 1635pp 63ndash71 1998
[45] A Montella L Imbriani and F Mauriello ldquoFactors Con-tributing to Run-off-the-Road Severe Crashesrdquo in Proceedingsof the Transportation Research Board 94th Annual MeetingWashington DC USA 2015
[46] A Oggero R M Darbra M Munoz E Planas and J Casal ldquoAsurveyof accidents occurring during the transport of hazardoussubstances by road and railrdquo Journal of Hazardous Materialsvol 133 no 1-3 pp 1ndash7 2006
[47] R Tay and S M Rifaat ldquoFactors contributing to the severity ofintersection crashesrdquo Journal of Advanced Transportation vol41 no 3 pp 245ndash265 2010
[48] Q Lin andW A Nixon ldquoEffects of Adverse Weather on TrafficCrashes Systematic Review andMeta-AnalysisrdquoTransportationResearch Record Journal of the Transportation Research Boardvol 2055 no 2055 pp 139ndash146 2008
[49] P Konstantopoulos P Chapman and D Crundall ldquoDriverrsquosvisual attention as a function of driving experience and visibil-ity Using a driving simulator to explore driversrsquo eye movementsin day night and rain drivingrdquo Accident Analysis amp Preventionvol 42 no 3 pp 827ndash834 2010
[50] H-D Zhang and X-P Zheng ldquoCharacteristics of hazardouschemical accidents in China A statistical investigationrdquo Journalof Loss Prevention in the Process Industries vol 25 no 4 pp686ndash693 2012
[51] J Pahukula S Hernandez and A Unnikrishnan ldquoA time ofday analysis of crashes involving large trucks in urban areasrdquoAccident Analysis amp Prevention vol 75 pp 155ndash163 2015
