Hindawi Publishing CorporationJournal of Computer Networks and CommunicationsVolume 2013 Article ID 723913 7 pageshttpdxdoiorg1011552013723913
Research ArticleEfficient Cluster Head Selection Algorithm for MANET
Khalid Hussain1 Abdul Hanan Abdullah1 Saleem Iqbal1
Khalid M Awan1 and Faraz Ahsan2
1 Faculty of Computing Universiti Teknologi Malaysia Skudai Johor Bahru 83100 Malaysia2 University Institute of Information Technology PMAS-ARID Agriculture University Rawalpindi 46300 Pakistan
Correspondence should be addressed to Abdul Hanan Abdullah hananutmmy
Received 27 May 2013 Accepted 8 October 2013
Academic Editor Heidi Steendam
Copyright copy 2013 Khalid Hussain et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
In mobile ad hoc network (MANET) cluster head selection is considered a gigantic challenge In wireless sensor network LEACHprotocol can be used to select cluster head on the bases of energy but it is still a dispute in mobil ad hoc networks and especiallywhen nodes are itinerant In this paper we proposed an efficient cluster head selection algorithm (ECHSA) for selection of thecluster head efficiently inMobile ad hoc networks We evaluate our proposed algorithm through simulation in OMNet++ as well ason test bed we experience the result according to our assumption For further evaluation we also compare our proposed protocolwith several other protocols like LEACH-C and consequences show perfection
1 Introduction
Cluster head (CH) election is the process to select a nodewithin the cluster as a leader node Cluster Head main-tains the information related to its cluster This informationincludes a list of nodes in the cluster and the path to everynode [1]
The responsibility of the CH is to communicate with allthe nodes of its own cluster However CH must be able tocommunicate with the nodes of other clusters as well whichcan be directly or through the respective CH or throughgateways Communication is done in three steps First ofall the cluster head receives the data sent by its memberssecondly it compresses the data and finally transmits the datato the base station or other CH Suitable cluster head canreduce energy utilization and enhances the network lifetime[2]
Electing a specific node as a cluster head is a very impor-tant but sophisticated job Various factors can be consideredfor electing the best node as a cluster head [3] Some of thesefactors include location of the node with respect to othernodes mobility energy trust and throughput of the node
Nodes of WSN and MANET have limited battery andresources Process of election increases the overall processing
overhead of the network So the election process must alsoconsider the processing and energy limitations of the nodes
One cluster head per cluster must be selected duringan election process because multiple cluster heads within asingle cluster can give rise to cluster reformation Quality ofService (QoS) and routing management issues [4]
In the recent years various surveys of CH electionschemes were presented Aim of these surveys is to discusstheir parameters need of reclustering [5] and performance[6] However to the best of our knowledge no overview ofthe CH election emphasizing position of node in cluster trustfactor of nodes and single cluster head selection per electionprocess has been discussed so far
In this paper efforts have been made to discuss an exten-sive number of schemes proposed previously for CH electionin both WSN and MANET To have a better understandingcomparison of various CH selection techniques is made interms of parameters used and possibility of multiple CHsselection
By using the predistinct techniques spanning tree [4] todesignate the new cluster head registration authenticationand verification of each register node in such cluster aretime incontrollable assignment It can be practicable in wiredsituation where resource and time are not an immense
2 Journal of Computer Networks and Communications
concern But in tangible period and wireless ad hoc networkit is a contest
In this paper we proposed an Efficient Artificial Intelli-gence based algorithm for cluster head section in mobile adhoc network To evaluate the proposed algorithm we carriedout a two-dimensional domain simulation in OmNet++which provides conventional wireless scenarios to implementas shown in Figure 1 We use the blacklist mechanism whichis discussed in Section 2 the up-dation of the routing table isalso discussed in the subsectionThe new extended algorithmandnode authentication algorithm are discussed in Section 3Section 4 will show the simulation results Conclusion of thework along with some limitation and future work has beendiscussed in the last section
11 Cluster Head Selection in MANET All the abovemen-tioned research works are in the scenario of wireless sensornetworks WSN and MANET have some common featureslike limited battery mobility issues and so forth Howeverapplications of the WSN are not applicable in MANETbecause nodes inWSN are designed to sense data and send tothe central authority however in ad hoc network nodes mayhave complete processing capabilities for example laptopscell phones and so forth WSN has a central authority calledbase station whereas MANET is a completely independentnetwork without any infrastructure These differences raisethe need of some other solutions for election processes thatare specially designed for ad hoc networks
Weighed cluster algorithm (WCA) is proposed forMANET [7] WCA elects the CH based on factors likemobility ability to handle nodes communication range andso forth The algorithm calculates the average weight of eachnode based on the provided factors The node with theminimum weight is selected as a cluster head
In K-hop connectivity ID clustering algorithm (KCO-NID) [8] the node having maximum connectivity is electedas CH If two nodes have the same connectivity valuethen they select the node having lower ID as CH Anotherapproach is used for dynamic CH election based on energylevel of the node [9] In this approach nodes share their IDsand energy value using broadcast messages After randomperiod the node with maximum energy level will be electedas Cluster Head If two nodes have the same energy levelthe node having the maximum number of neighbors will beelected as Cluster Head
An identifier based clustering algorithm is proposed[10] In this scheme a unique ID is assigned to each nodeThe node having minimum ID is elected as cluster headDegree of a node is calculated by every node on the basis ofdistance parameter If the Euclidean distance [11] is within thetransmission range the node will be elected as CH
Two variants of the cluster head selection distance-constrained and size-constrained are proposed for MANET[2] Two different algorithms are proposed for cluster headelection First algorithm is based on distance According tothis algorithm CH is selected if every member node is withina limited distance from the nearest CH Second algorithm isbased on the size of the cluster where each cluster is onlyallowed to have a limited number of members In this case
Figure 1 Conventional wireless network
CH is selected such that the size of each cluster is not largerthan a predefined value
Another solution for CH election is proposed forMANET [12] In this paper authors proposed an adaptiveinvoked weighted clustering algorithm which maintainsstable clusters In Weighted Clustering Algorithm (WCA)a node is selected to be the cluster head with minimumweighted sum of four indices-node degree (number of directlinks to its neighbors) sum of distances to all its neighboringnodes mobility and remaining battery power respectively[7]WCA lacks in knowing the weights of all the nodes beforestarting the clustering process and in drainingCHs rapidly Tosolve this problem S Rouhini proposed a probability basedadaptive invoked weighted clustering algorithm (PAIWCA)This can enhance the stability of the network by taking batterypower of the node into consideration for selecting clusterheads and for forming clusters The weight of a node iscalculated before the clustering process thus by minimizingthe overhead of reclustering in electing a cluster head
Reputation-based trust management strategy for clus-tered ad hoc networks is proposed for clustered ad hocnetworks [13] In this paper a cluster head backupmechanismwas maintained The existing CH selects its backup whichhas maximum trust value Cluster head updates all theinformation to its backup If CH cannot communicate withother nodes it transfers this role to the backup CH
Another trust based approach is proposed for MANET[14] In this work any candidate for CH broadcasts themessage with its mobility battery power value to all itsone hop neighbors Receivers calculate the global weight ofthe sender by using the received information and addingtrust value of the sender If global weight is greater than apredefined value the receiver will vote for the sender Aftera certain time the candidate node will count the votes Ifthe number of votes is greater than half of the number ofmembers it advertises itself as leader
We have compared the abovementioned techniques inTable 2 In this table we compared the techniques and high-lighted the parameters used in the above solutions We alsohighlight the handling of case of tie in the above mentionedalgorithms
12 Cluster-Weighed Modeling According to the statisticscluster-weighted modeling (CWM) is an algorithm-based
Journal of Computer Networks and Communications 3
method for analyzing the nonlinear prediction of outputs(reliant variables) from inputs (liberated variables) con-structed on density estimation by a conventional of simu-lations (clusters) that are every theoretically suitable in asubsection of the input galaxy The inclusive methodologyworks conjointly with input-output galaxy and an originalstyle was suggested by Neil Gershenfeld [3 15]
13 Basic Form of Model To construct the cluster model onthe basis of input delinquent output can be formulated like119910 = 119909 + 119890
1015840 where 1198901015840 being the error for packet mishandlingretransmissions To achieve the expected theory on the basisof the output variable 119910 in reflection of input variable 119909the joint probability solidity function can be explained as(119910 119909) In this situation the input and output variables canbe invariant or multivariate For the appropriateness anytypical constraints are not signposted in the symbolizationhere and numerous changed behaviors of these are probableincluding backdrop of immobile values as a stride in thestandardization or are considered expanding via Bayesiananalysis The essential prophesied tenets are acquired byfabricating the conditional probability solidity (119910 | 119909) fromwhich the calculation using the restrictive estimated valuecan be acquired with the restrictive modification providing asymptom of ambiguity
The significant step of the demonstrating is that 119901(119910 | 119909)is presumed to yield the following procedure as a combina-tion model
119901 (119910 119909) =
119899
sum
1
119908119895119901119895(119910 119909) (1)
where 119899 is the number of clusters and 119908119895 are weight (total
number of packets sent from source to destination in specifictime) that total to oneTheoccupations 119901
119895(119910 119909) are common
probability solidity functions that communicate to each of the119899 clustersThese functions are exhibited by disintegration intoa conditional and a peripheral solidity
119901119895(119910 119909) = 119901
119895(119910 | 119909) 119901
119895(119909) (2)
where 119901119895(119910 119909) is a successful packet delivery expecting 119910
assumed 119909 and it is assumed that the input-output coupleshould be associated with node 119895 on the source of theassessment of 119909 This typical might be a waning archetypalin the weakest circumstances
119901119895(119909) is imperiously solidity for tenets of 119909 assuming that
the input-output couple should be concomitant with node119895 The qualified sizes of these utilities between the clustersconcludewhether a specific assessment of119909 is connectedwithany assumed cluster center This solidity influence needs tobe a Gaussian function highlighted as a parameter signifyingthe cluster center
In identical fashion for regression analysis it will besignificant to deliberately renovate initial data as portionof the overall modeling strategy The potential candidatesneed to be evaluated modestly in an autonomous fashionwhile minimizing the possible errors of packet mishandlingfor each cluster on the basis of standard disseminationsincorporating the cluster-weighing densities 119901
119895(119909)
14 Technique Assume that 120579 is the set of ambiguous factorsand predictions in the perfect Assume that 119864 is taken asthe altered proof Before evaluating the assumed result onthe bases of preliminary previous probability distribution weassume that the confirmation is occupied into justification toassume about 120579
To evaluate our proposed technique Baysrsquo theorem isapplied
119875(120579 | 119864) is the probability distribution of the ambiguousamounts and subsequently the confirmation is reserved intointerpretation the posterior probability
119875(120579) is the probability distribution in lieu of ambiguityroughly and the factors and expectations formerly and theindication is reserved into interpretation the prior probabil-ity 119875(119864 | 120579)119875(119864) is a factor in lieu of the impression of theindication on conclusions about 120579
On the other hand in the sustenance of Bayesrsquo theorempossibly will be applied continually It is continuous practicein which every application the last one posterior becomes thedifferent preceding
2 Explanation
21 Elements Explanation 120579 used as a special case for theinterpretation of the influence and it delineates a discrete setof standards Assume that 119867 is one of these potential stan-dards In the following equation 119867 represents ldquohypothesisrdquootherwise usually 119867 epitomizes indeterminate constraint ormagnitude in a perfect
When the confirmation becomes according to the suggestedassumption it seems and gives more confidence whenhypothesis is true On the other side the antithesis disputerelates for a diminution in confidence In this circumstanceconfidence does not modify
22 Bayes Estimator As per estimation philosophy anddecision philosophy a Bayes estimator or a Bayes action isa measuring and authentic methodology which diminishesthe posterior predictable assessment on the bases of a lossfunction which calls posterior expected loss On the otherside it enhances the posterior probability of an effective task
23 Description Understand that an anonymous constraint 120579is recognized to have an earlier dissemination 120587 Let 120575 = 120575(119909)be an estimator of 120579 (constructed on certain capacities119909) andlet 119871(120579 120575) be a harm task such as adjusted inaccuracy TheBayes risk of 120575 is demarcated as 120587119871(120579 120575) and someplace theanticipation occupied terminated the probability distributionof 120579 this explains the threat occupation as a task of 120575
4 Journal of Computer Networks and Communications
Table 1 Simulation parameters for cluster head selection
Examined protocol AODVSimulation time 25minTransmission range 250mTraffic type UDPTraffic load 255 250 245 ppsPacket size 4096Data rate trunc-normalChannel error rate 00Channel data rate 1104858119890 + 6
An estimator 120575 is supposed to be a Bayes estimator if it dimin-ishes theBayes threat between all estimators Consistently theestimator which diminishes the subsequent predictable harm119864119871(120579 120575) | 119909 for each 119909 also diminishes the Bayes threat andconsequently is a Bayes estimator [1]
If the preceding is inappropriate then an estimator whichdiminishes the subsequent predictable damage for each 119909 iscalled a generalized Bayes estimator [2]
3 Proposed Solution
31 Black andWhite List In addition to two additional fieldsidentified earlier that have been inculcated in the X-AODVfor the purpose of decision making of routing path anotherfield ldquoBlackNwhiterdquo is added that maintains the status ofeach node based on malicious activity Figure 2 shows snapof scenario in which by using the X-AODV the neighbornode identified node 11 as malicious X-AODV change thecolor of malicious node to gray To fairly evaluate we runthe simulation for a period of 25 sec in multiple networkscenarios that is 12 15 25 40 50 and 65 nodes Duringthis period our proposed protocol identified the maliciousbehavior of the nodes shown in Table 2
4 Simulation Result and Analysis
According to defined parameters in Table 1 we create threescenarios but in base parameters theywere the same as aboveIn every scenario we change the traffic load and evaluate theperformance of the X-AODV In our proposed protocol wehave not chosen the predefined cluster head nor attack ormalicious node in the network For crystal evaluation we runevery scenario for a period of 25 minutes and during thisperiod of time our proposed protocol detects the maliciousbehavior of the multiple nodes In Table 2 we presented the20-node scenario in which X-AODV detect multiple node asmalicious
Figures 2(a) and 2(b) show the three- and five-minute realsimulation picture of the 10-node scenario which shows thatECHSA detect node 5 just after 3 minutes and nine and twojust after five minutes as malicious andmark the node as grayand set its flag in routing table as shown in Table 2
Cluster Head selection is usually based on spanningtree that works on sequence number and the node withminimum sequence number is selected as Cluster Head In
our proposed solution Cluster Head selectionreselection isperformed periodically But instead of just using the conceptof spanning tree it also takes into consideration the Blackamp White while selecting Cluster head Any node havingminimum sequence number but with status of black is notselected as Cluster Head instead chance is given to the nexthigher sequence number node Figure 3 shows the scenarioof selecting Cluster Head highlighted as yellow
In Table 2 we have defined the 20-node experimentalresults in which the ECHSA make an election within thosegood nodes which have the status of 0 Due to the segregationprocess the good nodes have comparatively less quantity soelection process takes less amount of resources
Algorithm 1 describes the overall mechanism for selec-tion of the ClusterHeadWhen a network is required to selectthe cluster head then every node will check its routing tableaccording to Algorithm 1 The mathematical representationof X-AODV with probabilistic extension along with theparameters describe in Notations section
When a node is elected as Cluster Head it is mandatoryfor that node to inform all registered nodes in the clusterabout its selection and the register node required to revoketheir authorization certificate from new cluster head Energyis a big issue inWirelessNetwork as well as in sensor networkso we also minimize this process in our proposed protocolAlgorithm 2 describes the Cluster Head announcement andalso the acceptance from the registration node as well asthe authorization from the Cluster Head In this processwe accommodate this announcement and certificate revokeprocedure through these nonce messages
Journal of Computer Networks and Communications 5
6 4
7
095
1
2 3
8
(a)
6 4
7
095
1
2 3
8
(b)
Figure 2 (a) and (b) Disqualify node identification for cluster head selection process (blackwhitelist)
Given a vector 120579 of parameters to determine a prior PDF119901(120579) over those parameters and a PDF 119901(119910 | 120579 120585) for makingobservation 119910 given parameter values 120579 and an experimentdesign 120585 the posterior PDF can be calculated using Bayesrsquotheorem
where 119880(119910 120585) is some real-valued functional of the posteriorPDF 119901(120579 | 119910 120585) after making observation y using an exper-iment design 120585
To evaluate the Cluster Performance we use the sameparameters described in Table 1 but in three different scenar-ios
Packet loss ratio is one of the important parameters fora node to be considered a good node as well as considerfor Cluster Head selection In these scenarios on the basesof throughput we evaluate the performance of X-AODV InFigure 4 we evaluate the performance of X-AODV just on 10nodes after a period of time we increase the number of nodesup to 65 and then reevaluate the performance of the proposedprotocol Figure 5 presents the performance evaluation of
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
concern But in tangible period and wireless ad hoc networkit is a contest
In this paper we proposed an Efficient Artificial Intelli-gence based algorithm for cluster head section in mobile adhoc network To evaluate the proposed algorithm we carriedout a two-dimensional domain simulation in OmNet++which provides conventional wireless scenarios to implementas shown in Figure 1 We use the blacklist mechanism whichis discussed in Section 2 the up-dation of the routing table isalso discussed in the subsectionThe new extended algorithmandnode authentication algorithm are discussed in Section 3Section 4 will show the simulation results Conclusion of thework along with some limitation and future work has beendiscussed in the last section
11 Cluster Head Selection in MANET All the abovemen-tioned research works are in the scenario of wireless sensornetworks WSN and MANET have some common featureslike limited battery mobility issues and so forth Howeverapplications of the WSN are not applicable in MANETbecause nodes inWSN are designed to sense data and send tothe central authority however in ad hoc network nodes mayhave complete processing capabilities for example laptopscell phones and so forth WSN has a central authority calledbase station whereas MANET is a completely independentnetwork without any infrastructure These differences raisethe need of some other solutions for election processes thatare specially designed for ad hoc networks
Weighed cluster algorithm (WCA) is proposed forMANET [7] WCA elects the CH based on factors likemobility ability to handle nodes communication range andso forth The algorithm calculates the average weight of eachnode based on the provided factors The node with theminimum weight is selected as a cluster head
In K-hop connectivity ID clustering algorithm (KCO-NID) [8] the node having maximum connectivity is electedas CH If two nodes have the same connectivity valuethen they select the node having lower ID as CH Anotherapproach is used for dynamic CH election based on energylevel of the node [9] In this approach nodes share their IDsand energy value using broadcast messages After randomperiod the node with maximum energy level will be electedas Cluster Head If two nodes have the same energy levelthe node having the maximum number of neighbors will beelected as Cluster Head
An identifier based clustering algorithm is proposed[10] In this scheme a unique ID is assigned to each nodeThe node having minimum ID is elected as cluster headDegree of a node is calculated by every node on the basis ofdistance parameter If the Euclidean distance [11] is within thetransmission range the node will be elected as CH
Two variants of the cluster head selection distance-constrained and size-constrained are proposed for MANET[2] Two different algorithms are proposed for cluster headelection First algorithm is based on distance According tothis algorithm CH is selected if every member node is withina limited distance from the nearest CH Second algorithm isbased on the size of the cluster where each cluster is onlyallowed to have a limited number of members In this case
Figure 1 Conventional wireless network
CH is selected such that the size of each cluster is not largerthan a predefined value
Another solution for CH election is proposed forMANET [12] In this paper authors proposed an adaptiveinvoked weighted clustering algorithm which maintainsstable clusters In Weighted Clustering Algorithm (WCA)a node is selected to be the cluster head with minimumweighted sum of four indices-node degree (number of directlinks to its neighbors) sum of distances to all its neighboringnodes mobility and remaining battery power respectively[7]WCA lacks in knowing the weights of all the nodes beforestarting the clustering process and in drainingCHs rapidly Tosolve this problem S Rouhini proposed a probability basedadaptive invoked weighted clustering algorithm (PAIWCA)This can enhance the stability of the network by taking batterypower of the node into consideration for selecting clusterheads and for forming clusters The weight of a node iscalculated before the clustering process thus by minimizingthe overhead of reclustering in electing a cluster head
Reputation-based trust management strategy for clus-tered ad hoc networks is proposed for clustered ad hocnetworks [13] In this paper a cluster head backupmechanismwas maintained The existing CH selects its backup whichhas maximum trust value Cluster head updates all theinformation to its backup If CH cannot communicate withother nodes it transfers this role to the backup CH
Another trust based approach is proposed for MANET[14] In this work any candidate for CH broadcasts themessage with its mobility battery power value to all itsone hop neighbors Receivers calculate the global weight ofthe sender by using the received information and addingtrust value of the sender If global weight is greater than apredefined value the receiver will vote for the sender Aftera certain time the candidate node will count the votes Ifthe number of votes is greater than half of the number ofmembers it advertises itself as leader
We have compared the abovementioned techniques inTable 2 In this table we compared the techniques and high-lighted the parameters used in the above solutions We alsohighlight the handling of case of tie in the above mentionedalgorithms
12 Cluster-Weighed Modeling According to the statisticscluster-weighted modeling (CWM) is an algorithm-based
Journal of Computer Networks and Communications 3
method for analyzing the nonlinear prediction of outputs(reliant variables) from inputs (liberated variables) con-structed on density estimation by a conventional of simu-lations (clusters) that are every theoretically suitable in asubsection of the input galaxy The inclusive methodologyworks conjointly with input-output galaxy and an originalstyle was suggested by Neil Gershenfeld [3 15]
13 Basic Form of Model To construct the cluster model onthe basis of input delinquent output can be formulated like119910 = 119909 + 119890
1015840 where 1198901015840 being the error for packet mishandlingretransmissions To achieve the expected theory on the basisof the output variable 119910 in reflection of input variable 119909the joint probability solidity function can be explained as(119910 119909) In this situation the input and output variables canbe invariant or multivariate For the appropriateness anytypical constraints are not signposted in the symbolizationhere and numerous changed behaviors of these are probableincluding backdrop of immobile values as a stride in thestandardization or are considered expanding via Bayesiananalysis The essential prophesied tenets are acquired byfabricating the conditional probability solidity (119910 | 119909) fromwhich the calculation using the restrictive estimated valuecan be acquired with the restrictive modification providing asymptom of ambiguity
The significant step of the demonstrating is that 119901(119910 | 119909)is presumed to yield the following procedure as a combina-tion model
119901 (119910 119909) =
119899
sum
1
119908119895119901119895(119910 119909) (1)
where 119899 is the number of clusters and 119908119895 are weight (total
number of packets sent from source to destination in specifictime) that total to oneTheoccupations 119901
119895(119910 119909) are common
probability solidity functions that communicate to each of the119899 clustersThese functions are exhibited by disintegration intoa conditional and a peripheral solidity
119901119895(119910 119909) = 119901
119895(119910 | 119909) 119901
119895(119909) (2)
where 119901119895(119910 119909) is a successful packet delivery expecting 119910
assumed 119909 and it is assumed that the input-output coupleshould be associated with node 119895 on the source of theassessment of 119909 This typical might be a waning archetypalin the weakest circumstances
119901119895(119909) is imperiously solidity for tenets of 119909 assuming that
the input-output couple should be concomitant with node119895 The qualified sizes of these utilities between the clustersconcludewhether a specific assessment of119909 is connectedwithany assumed cluster center This solidity influence needs tobe a Gaussian function highlighted as a parameter signifyingthe cluster center
In identical fashion for regression analysis it will besignificant to deliberately renovate initial data as portionof the overall modeling strategy The potential candidatesneed to be evaluated modestly in an autonomous fashionwhile minimizing the possible errors of packet mishandlingfor each cluster on the basis of standard disseminationsincorporating the cluster-weighing densities 119901
119895(119909)
14 Technique Assume that 120579 is the set of ambiguous factorsand predictions in the perfect Assume that 119864 is taken asthe altered proof Before evaluating the assumed result onthe bases of preliminary previous probability distribution weassume that the confirmation is occupied into justification toassume about 120579
To evaluate our proposed technique Baysrsquo theorem isapplied
119875(120579 | 119864) is the probability distribution of the ambiguousamounts and subsequently the confirmation is reserved intointerpretation the posterior probability
119875(120579) is the probability distribution in lieu of ambiguityroughly and the factors and expectations formerly and theindication is reserved into interpretation the prior probabil-ity 119875(119864 | 120579)119875(119864) is a factor in lieu of the impression of theindication on conclusions about 120579
On the other hand in the sustenance of Bayesrsquo theorempossibly will be applied continually It is continuous practicein which every application the last one posterior becomes thedifferent preceding
2 Explanation
21 Elements Explanation 120579 used as a special case for theinterpretation of the influence and it delineates a discrete setof standards Assume that 119867 is one of these potential stan-dards In the following equation 119867 represents ldquohypothesisrdquootherwise usually 119867 epitomizes indeterminate constraint ormagnitude in a perfect
When the confirmation becomes according to the suggestedassumption it seems and gives more confidence whenhypothesis is true On the other side the antithesis disputerelates for a diminution in confidence In this circumstanceconfidence does not modify
22 Bayes Estimator As per estimation philosophy anddecision philosophy a Bayes estimator or a Bayes action isa measuring and authentic methodology which diminishesthe posterior predictable assessment on the bases of a lossfunction which calls posterior expected loss On the otherside it enhances the posterior probability of an effective task
23 Description Understand that an anonymous constraint 120579is recognized to have an earlier dissemination 120587 Let 120575 = 120575(119909)be an estimator of 120579 (constructed on certain capacities119909) andlet 119871(120579 120575) be a harm task such as adjusted inaccuracy TheBayes risk of 120575 is demarcated as 120587119871(120579 120575) and someplace theanticipation occupied terminated the probability distributionof 120579 this explains the threat occupation as a task of 120575
4 Journal of Computer Networks and Communications
Table 1 Simulation parameters for cluster head selection
Examined protocol AODVSimulation time 25minTransmission range 250mTraffic type UDPTraffic load 255 250 245 ppsPacket size 4096Data rate trunc-normalChannel error rate 00Channel data rate 1104858119890 + 6
An estimator 120575 is supposed to be a Bayes estimator if it dimin-ishes theBayes threat between all estimators Consistently theestimator which diminishes the subsequent predictable harm119864119871(120579 120575) | 119909 for each 119909 also diminishes the Bayes threat andconsequently is a Bayes estimator [1]
If the preceding is inappropriate then an estimator whichdiminishes the subsequent predictable damage for each 119909 iscalled a generalized Bayes estimator [2]
3 Proposed Solution
31 Black andWhite List In addition to two additional fieldsidentified earlier that have been inculcated in the X-AODVfor the purpose of decision making of routing path anotherfield ldquoBlackNwhiterdquo is added that maintains the status ofeach node based on malicious activity Figure 2 shows snapof scenario in which by using the X-AODV the neighbornode identified node 11 as malicious X-AODV change thecolor of malicious node to gray To fairly evaluate we runthe simulation for a period of 25 sec in multiple networkscenarios that is 12 15 25 40 50 and 65 nodes Duringthis period our proposed protocol identified the maliciousbehavior of the nodes shown in Table 2
4 Simulation Result and Analysis
According to defined parameters in Table 1 we create threescenarios but in base parameters theywere the same as aboveIn every scenario we change the traffic load and evaluate theperformance of the X-AODV In our proposed protocol wehave not chosen the predefined cluster head nor attack ormalicious node in the network For crystal evaluation we runevery scenario for a period of 25 minutes and during thisperiod of time our proposed protocol detects the maliciousbehavior of the multiple nodes In Table 2 we presented the20-node scenario in which X-AODV detect multiple node asmalicious
Figures 2(a) and 2(b) show the three- and five-minute realsimulation picture of the 10-node scenario which shows thatECHSA detect node 5 just after 3 minutes and nine and twojust after five minutes as malicious andmark the node as grayand set its flag in routing table as shown in Table 2
Cluster Head selection is usually based on spanningtree that works on sequence number and the node withminimum sequence number is selected as Cluster Head In
our proposed solution Cluster Head selectionreselection isperformed periodically But instead of just using the conceptof spanning tree it also takes into consideration the Blackamp White while selecting Cluster head Any node havingminimum sequence number but with status of black is notselected as Cluster Head instead chance is given to the nexthigher sequence number node Figure 3 shows the scenarioof selecting Cluster Head highlighted as yellow
In Table 2 we have defined the 20-node experimentalresults in which the ECHSA make an election within thosegood nodes which have the status of 0 Due to the segregationprocess the good nodes have comparatively less quantity soelection process takes less amount of resources
Algorithm 1 describes the overall mechanism for selec-tion of the ClusterHeadWhen a network is required to selectthe cluster head then every node will check its routing tableaccording to Algorithm 1 The mathematical representationof X-AODV with probabilistic extension along with theparameters describe in Notations section
When a node is elected as Cluster Head it is mandatoryfor that node to inform all registered nodes in the clusterabout its selection and the register node required to revoketheir authorization certificate from new cluster head Energyis a big issue inWirelessNetwork as well as in sensor networkso we also minimize this process in our proposed protocolAlgorithm 2 describes the Cluster Head announcement andalso the acceptance from the registration node as well asthe authorization from the Cluster Head In this processwe accommodate this announcement and certificate revokeprocedure through these nonce messages
Journal of Computer Networks and Communications 5
6 4
7
095
1
2 3
8
(a)
6 4
7
095
1
2 3
8
(b)
Figure 2 (a) and (b) Disqualify node identification for cluster head selection process (blackwhitelist)
Given a vector 120579 of parameters to determine a prior PDF119901(120579) over those parameters and a PDF 119901(119910 | 120579 120585) for makingobservation 119910 given parameter values 120579 and an experimentdesign 120585 the posterior PDF can be calculated using Bayesrsquotheorem
where 119880(119910 120585) is some real-valued functional of the posteriorPDF 119901(120579 | 119910 120585) after making observation y using an exper-iment design 120585
To evaluate the Cluster Performance we use the sameparameters described in Table 1 but in three different scenar-ios
Packet loss ratio is one of the important parameters fora node to be considered a good node as well as considerfor Cluster Head selection In these scenarios on the basesof throughput we evaluate the performance of X-AODV InFigure 4 we evaluate the performance of X-AODV just on 10nodes after a period of time we increase the number of nodesup to 65 and then reevaluate the performance of the proposedprotocol Figure 5 presents the performance evaluation of
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
method for analyzing the nonlinear prediction of outputs(reliant variables) from inputs (liberated variables) con-structed on density estimation by a conventional of simu-lations (clusters) that are every theoretically suitable in asubsection of the input galaxy The inclusive methodologyworks conjointly with input-output galaxy and an originalstyle was suggested by Neil Gershenfeld [3 15]
13 Basic Form of Model To construct the cluster model onthe basis of input delinquent output can be formulated like119910 = 119909 + 119890
1015840 where 1198901015840 being the error for packet mishandlingretransmissions To achieve the expected theory on the basisof the output variable 119910 in reflection of input variable 119909the joint probability solidity function can be explained as(119910 119909) In this situation the input and output variables canbe invariant or multivariate For the appropriateness anytypical constraints are not signposted in the symbolizationhere and numerous changed behaviors of these are probableincluding backdrop of immobile values as a stride in thestandardization or are considered expanding via Bayesiananalysis The essential prophesied tenets are acquired byfabricating the conditional probability solidity (119910 | 119909) fromwhich the calculation using the restrictive estimated valuecan be acquired with the restrictive modification providing asymptom of ambiguity
The significant step of the demonstrating is that 119901(119910 | 119909)is presumed to yield the following procedure as a combina-tion model
119901 (119910 119909) =
119899
sum
1
119908119895119901119895(119910 119909) (1)
where 119899 is the number of clusters and 119908119895 are weight (total
number of packets sent from source to destination in specifictime) that total to oneTheoccupations 119901
119895(119910 119909) are common
probability solidity functions that communicate to each of the119899 clustersThese functions are exhibited by disintegration intoa conditional and a peripheral solidity
119901119895(119910 119909) = 119901
119895(119910 | 119909) 119901
119895(119909) (2)
where 119901119895(119910 119909) is a successful packet delivery expecting 119910
assumed 119909 and it is assumed that the input-output coupleshould be associated with node 119895 on the source of theassessment of 119909 This typical might be a waning archetypalin the weakest circumstances
119901119895(119909) is imperiously solidity for tenets of 119909 assuming that
the input-output couple should be concomitant with node119895 The qualified sizes of these utilities between the clustersconcludewhether a specific assessment of119909 is connectedwithany assumed cluster center This solidity influence needs tobe a Gaussian function highlighted as a parameter signifyingthe cluster center
In identical fashion for regression analysis it will besignificant to deliberately renovate initial data as portionof the overall modeling strategy The potential candidatesneed to be evaluated modestly in an autonomous fashionwhile minimizing the possible errors of packet mishandlingfor each cluster on the basis of standard disseminationsincorporating the cluster-weighing densities 119901
119895(119909)
14 Technique Assume that 120579 is the set of ambiguous factorsand predictions in the perfect Assume that 119864 is taken asthe altered proof Before evaluating the assumed result onthe bases of preliminary previous probability distribution weassume that the confirmation is occupied into justification toassume about 120579
To evaluate our proposed technique Baysrsquo theorem isapplied
119875(120579 | 119864) is the probability distribution of the ambiguousamounts and subsequently the confirmation is reserved intointerpretation the posterior probability
119875(120579) is the probability distribution in lieu of ambiguityroughly and the factors and expectations formerly and theindication is reserved into interpretation the prior probabil-ity 119875(119864 | 120579)119875(119864) is a factor in lieu of the impression of theindication on conclusions about 120579
On the other hand in the sustenance of Bayesrsquo theorempossibly will be applied continually It is continuous practicein which every application the last one posterior becomes thedifferent preceding
2 Explanation
21 Elements Explanation 120579 used as a special case for theinterpretation of the influence and it delineates a discrete setof standards Assume that 119867 is one of these potential stan-dards In the following equation 119867 represents ldquohypothesisrdquootherwise usually 119867 epitomizes indeterminate constraint ormagnitude in a perfect
When the confirmation becomes according to the suggestedassumption it seems and gives more confidence whenhypothesis is true On the other side the antithesis disputerelates for a diminution in confidence In this circumstanceconfidence does not modify
22 Bayes Estimator As per estimation philosophy anddecision philosophy a Bayes estimator or a Bayes action isa measuring and authentic methodology which diminishesthe posterior predictable assessment on the bases of a lossfunction which calls posterior expected loss On the otherside it enhances the posterior probability of an effective task
23 Description Understand that an anonymous constraint 120579is recognized to have an earlier dissemination 120587 Let 120575 = 120575(119909)be an estimator of 120579 (constructed on certain capacities119909) andlet 119871(120579 120575) be a harm task such as adjusted inaccuracy TheBayes risk of 120575 is demarcated as 120587119871(120579 120575) and someplace theanticipation occupied terminated the probability distributionof 120579 this explains the threat occupation as a task of 120575
4 Journal of Computer Networks and Communications
Table 1 Simulation parameters for cluster head selection
Examined protocol AODVSimulation time 25minTransmission range 250mTraffic type UDPTraffic load 255 250 245 ppsPacket size 4096Data rate trunc-normalChannel error rate 00Channel data rate 1104858119890 + 6
An estimator 120575 is supposed to be a Bayes estimator if it dimin-ishes theBayes threat between all estimators Consistently theestimator which diminishes the subsequent predictable harm119864119871(120579 120575) | 119909 for each 119909 also diminishes the Bayes threat andconsequently is a Bayes estimator [1]
If the preceding is inappropriate then an estimator whichdiminishes the subsequent predictable damage for each 119909 iscalled a generalized Bayes estimator [2]
3 Proposed Solution
31 Black andWhite List In addition to two additional fieldsidentified earlier that have been inculcated in the X-AODVfor the purpose of decision making of routing path anotherfield ldquoBlackNwhiterdquo is added that maintains the status ofeach node based on malicious activity Figure 2 shows snapof scenario in which by using the X-AODV the neighbornode identified node 11 as malicious X-AODV change thecolor of malicious node to gray To fairly evaluate we runthe simulation for a period of 25 sec in multiple networkscenarios that is 12 15 25 40 50 and 65 nodes Duringthis period our proposed protocol identified the maliciousbehavior of the nodes shown in Table 2
4 Simulation Result and Analysis
According to defined parameters in Table 1 we create threescenarios but in base parameters theywere the same as aboveIn every scenario we change the traffic load and evaluate theperformance of the X-AODV In our proposed protocol wehave not chosen the predefined cluster head nor attack ormalicious node in the network For crystal evaluation we runevery scenario for a period of 25 minutes and during thisperiod of time our proposed protocol detects the maliciousbehavior of the multiple nodes In Table 2 we presented the20-node scenario in which X-AODV detect multiple node asmalicious
Figures 2(a) and 2(b) show the three- and five-minute realsimulation picture of the 10-node scenario which shows thatECHSA detect node 5 just after 3 minutes and nine and twojust after five minutes as malicious andmark the node as grayand set its flag in routing table as shown in Table 2
Cluster Head selection is usually based on spanningtree that works on sequence number and the node withminimum sequence number is selected as Cluster Head In
our proposed solution Cluster Head selectionreselection isperformed periodically But instead of just using the conceptof spanning tree it also takes into consideration the Blackamp White while selecting Cluster head Any node havingminimum sequence number but with status of black is notselected as Cluster Head instead chance is given to the nexthigher sequence number node Figure 3 shows the scenarioof selecting Cluster Head highlighted as yellow
In Table 2 we have defined the 20-node experimentalresults in which the ECHSA make an election within thosegood nodes which have the status of 0 Due to the segregationprocess the good nodes have comparatively less quantity soelection process takes less amount of resources
Algorithm 1 describes the overall mechanism for selec-tion of the ClusterHeadWhen a network is required to selectthe cluster head then every node will check its routing tableaccording to Algorithm 1 The mathematical representationof X-AODV with probabilistic extension along with theparameters describe in Notations section
When a node is elected as Cluster Head it is mandatoryfor that node to inform all registered nodes in the clusterabout its selection and the register node required to revoketheir authorization certificate from new cluster head Energyis a big issue inWirelessNetwork as well as in sensor networkso we also minimize this process in our proposed protocolAlgorithm 2 describes the Cluster Head announcement andalso the acceptance from the registration node as well asthe authorization from the Cluster Head In this processwe accommodate this announcement and certificate revokeprocedure through these nonce messages
Journal of Computer Networks and Communications 5
6 4
7
095
1
2 3
8
(a)
6 4
7
095
1
2 3
8
(b)
Figure 2 (a) and (b) Disqualify node identification for cluster head selection process (blackwhitelist)
Given a vector 120579 of parameters to determine a prior PDF119901(120579) over those parameters and a PDF 119901(119910 | 120579 120585) for makingobservation 119910 given parameter values 120579 and an experimentdesign 120585 the posterior PDF can be calculated using Bayesrsquotheorem
where 119880(119910 120585) is some real-valued functional of the posteriorPDF 119901(120579 | 119910 120585) after making observation y using an exper-iment design 120585
To evaluate the Cluster Performance we use the sameparameters described in Table 1 but in three different scenar-ios
Packet loss ratio is one of the important parameters fora node to be considered a good node as well as considerfor Cluster Head selection In these scenarios on the basesof throughput we evaluate the performance of X-AODV InFigure 4 we evaluate the performance of X-AODV just on 10nodes after a period of time we increase the number of nodesup to 65 and then reevaluate the performance of the proposedprotocol Figure 5 presents the performance evaluation of
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
Table 1 Simulation parameters for cluster head selection
Examined protocol AODVSimulation time 25minTransmission range 250mTraffic type UDPTraffic load 255 250 245 ppsPacket size 4096Data rate trunc-normalChannel error rate 00Channel data rate 1104858119890 + 6
An estimator 120575 is supposed to be a Bayes estimator if it dimin-ishes theBayes threat between all estimators Consistently theestimator which diminishes the subsequent predictable harm119864119871(120579 120575) | 119909 for each 119909 also diminishes the Bayes threat andconsequently is a Bayes estimator [1]
If the preceding is inappropriate then an estimator whichdiminishes the subsequent predictable damage for each 119909 iscalled a generalized Bayes estimator [2]
3 Proposed Solution
31 Black andWhite List In addition to two additional fieldsidentified earlier that have been inculcated in the X-AODVfor the purpose of decision making of routing path anotherfield ldquoBlackNwhiterdquo is added that maintains the status ofeach node based on malicious activity Figure 2 shows snapof scenario in which by using the X-AODV the neighbornode identified node 11 as malicious X-AODV change thecolor of malicious node to gray To fairly evaluate we runthe simulation for a period of 25 sec in multiple networkscenarios that is 12 15 25 40 50 and 65 nodes Duringthis period our proposed protocol identified the maliciousbehavior of the nodes shown in Table 2
4 Simulation Result and Analysis
According to defined parameters in Table 1 we create threescenarios but in base parameters theywere the same as aboveIn every scenario we change the traffic load and evaluate theperformance of the X-AODV In our proposed protocol wehave not chosen the predefined cluster head nor attack ormalicious node in the network For crystal evaluation we runevery scenario for a period of 25 minutes and during thisperiod of time our proposed protocol detects the maliciousbehavior of the multiple nodes In Table 2 we presented the20-node scenario in which X-AODV detect multiple node asmalicious
Figures 2(a) and 2(b) show the three- and five-minute realsimulation picture of the 10-node scenario which shows thatECHSA detect node 5 just after 3 minutes and nine and twojust after five minutes as malicious andmark the node as grayand set its flag in routing table as shown in Table 2
Cluster Head selection is usually based on spanningtree that works on sequence number and the node withminimum sequence number is selected as Cluster Head In
our proposed solution Cluster Head selectionreselection isperformed periodically But instead of just using the conceptof spanning tree it also takes into consideration the Blackamp White while selecting Cluster head Any node havingminimum sequence number but with status of black is notselected as Cluster Head instead chance is given to the nexthigher sequence number node Figure 3 shows the scenarioof selecting Cluster Head highlighted as yellow
In Table 2 we have defined the 20-node experimentalresults in which the ECHSA make an election within thosegood nodes which have the status of 0 Due to the segregationprocess the good nodes have comparatively less quantity soelection process takes less amount of resources
Algorithm 1 describes the overall mechanism for selec-tion of the ClusterHeadWhen a network is required to selectthe cluster head then every node will check its routing tableaccording to Algorithm 1 The mathematical representationof X-AODV with probabilistic extension along with theparameters describe in Notations section
When a node is elected as Cluster Head it is mandatoryfor that node to inform all registered nodes in the clusterabout its selection and the register node required to revoketheir authorization certificate from new cluster head Energyis a big issue inWirelessNetwork as well as in sensor networkso we also minimize this process in our proposed protocolAlgorithm 2 describes the Cluster Head announcement andalso the acceptance from the registration node as well asthe authorization from the Cluster Head In this processwe accommodate this announcement and certificate revokeprocedure through these nonce messages
Journal of Computer Networks and Communications 5
6 4
7
095
1
2 3
8
(a)
6 4
7
095
1
2 3
8
(b)
Figure 2 (a) and (b) Disqualify node identification for cluster head selection process (blackwhitelist)
Given a vector 120579 of parameters to determine a prior PDF119901(120579) over those parameters and a PDF 119901(119910 | 120579 120585) for makingobservation 119910 given parameter values 120579 and an experimentdesign 120585 the posterior PDF can be calculated using Bayesrsquotheorem
where 119880(119910 120585) is some real-valued functional of the posteriorPDF 119901(120579 | 119910 120585) after making observation y using an exper-iment design 120585
To evaluate the Cluster Performance we use the sameparameters described in Table 1 but in three different scenar-ios
Packet loss ratio is one of the important parameters fora node to be considered a good node as well as considerfor Cluster Head selection In these scenarios on the basesof throughput we evaluate the performance of X-AODV InFigure 4 we evaluate the performance of X-AODV just on 10nodes after a period of time we increase the number of nodesup to 65 and then reevaluate the performance of the proposedprotocol Figure 5 presents the performance evaluation of
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
Given a vector 120579 of parameters to determine a prior PDF119901(120579) over those parameters and a PDF 119901(119910 | 120579 120585) for makingobservation 119910 given parameter values 120579 and an experimentdesign 120585 the posterior PDF can be calculated using Bayesrsquotheorem
where 119880(119910 120585) is some real-valued functional of the posteriorPDF 119901(120579 | 119910 120585) after making observation y using an exper-iment design 120585
To evaluate the Cluster Performance we use the sameparameters described in Table 1 but in three different scenar-ios
Packet loss ratio is one of the important parameters fora node to be considered a good node as well as considerfor Cluster Head selection In these scenarios on the basesof throughput we evaluate the performance of X-AODV InFigure 4 we evaluate the performance of X-AODV just on 10nodes after a period of time we increase the number of nodesup to 65 and then reevaluate the performance of the proposedprotocol Figure 5 presents the performance evaluation of
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
20 nodes whereas in Figure 6 we compare our proposedprotocol with some latest protocols with the same parametersand found the Cluster Head Selection performance morebatter as compare to the previous one
Figure 4 describes the selection performance of ECHSAand in this scenario the coordination is between 15 nodesand with the help of IMBDM five nodes were detected asmalicious Hence at the time of CH selection those nodesare not allowed to participate for electionThen performanceof the network for the simulation time is shown in Figure 4Blue line shows the sent data whereas the brown line showsthe overall network throughput for the first minute wheremalicious nodes tried to inject fake routing entries on thenetwork to disrupt communicationThe result can be dividedinto 3 logical phases First while the default CHwas activatedat the time of network initialization a good throughput wasachieved among member nodes However as the maliciousnodes started disrupting the topology after approximately15 seconds the goodput of the network dropped to halfAround 40 seconds ECHSA was activated and blacklistednodes were detected and discarded for packet forwardinghence newCHwas selected and fresh routes were discoveredThus networkwas restored through new routes Similar is theinclination visible in the later part of the result Estimateddrop was around 20 which practically doubled due tomultiple malicious nodes in the network that is 5 Eventhen the network was able to maintain more than 50of communication in the first minute while the maliciousnodes were active on the control plane This shows thatthe proposed algorithm selects new cluster head withoutdisturbing the normal communication as well as requiredadditional resource
The same process was repeated to analyze the behaviorof ECHSA with increased member and malicious nodes Wekept the world size the same but doubled the number of goodand malicious nodes that is out of 30 nodes in total 10were malicious Consequently the election of a new CH washeld between 20 nodes The disruptions by malicious nodesand detection by good ones occurred little earlier mainlydue to increased nodes and network density However the
Figure 6 X-AODV comparison with different protocols
restoration of the network initiated on approximately thesame time the reason being delay in coordination betweenincreased nodes for CH and routes setupThe estimated dropwas around half (55) but overall 65 communication wassuccessful majorly due to more alternate routes possible asnetwork density increased
Based on the above scenarios it has been determined thatECHSA not only works in denser environment rather betterOn average ECHSAwhich has retained 63 communicationwith 13rd nodes within a network are malicious Above allthemalicious attackers on the control plane are discarded andnetwork starts functioning smoothly within the first minuteof the topology setup
At the end we also compare our proposed algorithmwith the other Cluster Head Selection protocol and Figure 6presents the comparison analysis with LEACH and LEACH-C
Journal of Computer Networks and Communications 7
5 Conclusion
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
In this paper we have presented a novel artificial intelligencebased Algorithm to select new cluster head in MANET Onthe bases of minimum packet loss ratio as well as maliciousbehavior of the node our algorithm excludes node for electionas cluster head ECHSA has the AI capabilities to select thecluster head by just populating theBampWlist Results and eval-uation show that our technique is more efficient and requiredminimum resource for cluster head selection With the helpof our proposed protocol a significant escalation comes inthe MANET lifetime By enhancing the AI capability (bayestimator) an additional enhancement in MANET lifetimeand resource consumption can be accomplished We alsoexperiment our algorithm in different scenarioswithmultipledata rate for critical evaluation and fair cluster head
Notations
Θ Parameters to be determined119884 Observation or data120585 Design119901(119910120579 120585) PDF for making observation 119910 given
parameter values 120579 and design 120585119901(120579) Prior PDF119901(119910120585) Marginal PDF in observation space119901(120579119910 120585) Posterior PDF119880(120585) Utility of the design 120585119880(119910 120585) Utility of the experiment outcome after
observation 119910 with design 120585
Acknowledgment
This research is supported by the Ministry of ScienceTechnology and Innovation (MOSTI) and was conducted incollaboration with the Research Management Center (RMC)at the Universiti Teknologi Malaysia (UTM) under Vot noRJ13000079284S014
References
[1] A A Abbasi andM Younis ldquoA survey on clustering algorithmsfor wireless sensor networksrdquo Computer Communications vol30 no 14-15 pp 2826ndash2841 2007
[2] R Agarwal and D Motwani ldquoSurvey of clustering algorithmsfor MANETrdquo httparxivorgabs09122303
[3] M Chatterjee S Sas and D Turgut ldquoAn on-demand weightedclustering algorithm (WCA) for ad hoc networksrdquo in Pro-ceedings of the IEEE Global Telecommunications Conference(GLOBECOM rsquo00) 2000
[4] P Chatterjee ldquoTrust based clustering and secure routing schemefor mobile ad hoc networksrdquo International Journal of ComputerNetworks and Communication vol 1 no 2 pp 84ndash97 2009
[5] S Chinara and S K Rath ldquoA survey on one-hop clusteringalgorithms in mobile ad hoc networksrdquo Journal of Network andSystems Management vol 17 no 1-2 pp 183ndash207 2009
[6] C-L Fok G-C Roman and C Lu ldquoRapid developmentand flexible deployment of adaptive wireless sensor networkapplicationsrdquo in Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems (ICDCS rsquo05) pp653ndash662 June 2005
[7] K Hussain A H Abdullah K M Awan F Ahsan and AHussain ldquoCluster head election schemes forWSN andMANETa surveyrdquoWorld Applied Sciences Journal vol 23 no 5 pp 611ndash620 2013
[8] D Nguyen P Minet T Kunz and L Lamont ldquoNew findingson the complexity of cluster head selection algorithmsrdquo inProceedings of the IEEE International Symposium on a World ofWirelessMobile andMultimediaNetworks (WoWMoM rsquo11) June2011
[9] F G Nocetti J S Gonzalez and I Stojmenovic ldquoConnectivitybased k-hop clustering in wireless networksrdquo Telecommunica-tion Systems vol 22 no 1ndash4 pp 205ndash220 2003
[10] K Ramesh and D K Somasundaram ldquoA comparative study ofclusterhead selection algorithms in wireless sensor networksrdquoInternational Journal of Computer ScienceampEngineering Surveyvol 2 no 4 2011
[11] S Rohini and K Indumathi ldquoProbability based adaptive invo-ked clustering algorithm in MANETsrdquo httparxivorgabs11021754
[12] L Xu andY Zhang ldquoA new reputation-based trustmanagementstrategy for clustered ad hoc networksrdquo in Proceedings of theInternational Conference on Networks Security Wireless Com-munications and Trusted Computing (NSWCTC rsquo09) pp 116ndash119 April 2009
[13] A Zabian A Ibrahim and F Al-Kalani ldquoDynamic head clusterelection algorithm for clustered Ad-Hoc networksrdquo Journal ofComputer Science vol 4 no 1 pp 42ndash50 2008
[14] N Zaman A B Abdullah and L T Jung ldquoOptimizationof energy usage in wireless sensor network using PositionResponsive Routing Protocol (PRRP)rdquo in Proceedings of theIEEE Symposium on Computers and Informatics (ISCI rsquo11) pp51ndash55 March 2011
[15] H S Lee K T Kim and H Y Youn ldquoA new cluster headselection scheme for long lifetime of wireless sensor networksrdquoin Computational Science and Its ApplicationsmdashICCSA 2006vol 3983 of Lecture Notes in Computer Science pp 519ndash528Springer 2006
![A modified cluster-head selection algorithm in …...data transmission [8]. In response to these problems, this paper presents a modified cluster-head selection algorithm based on](https://static.documents.pub/doc/80x56/5f081b087e708231d4205d43/a-modified-cluster-head-selection-algorithm-in-data-transmission-8-in-response.jpg)
