+ All Categories
Home > Documents > Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue...

Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue...

Date post: 20-May-2020
Category:
Upload: others
View: 3 times
Download: 0 times
Share this document with a friend
6
Revenue Oriented Air Quality Prediction MicroServices for Smart Cities Shajulin Benedict Indian Institute of Information Technology Kottayam, (Earlier - Guest Professor, TUM – Germany), Email: [email protected] Abstract—A notion of creating smart cities with integrated high end ICT technologies, including cloud, edge, IoTs, and so forth, has increased recently in order to provide a reasonably high comfort to city residents with limited human intervention. In fact, one of the major challenges faced by city authorities of developing countries is to control air pollution, which might cause chronic diseases and financial concerns to the residents. Although a few air quality monitoring and air pollution control measures exist, these solutions are not remaining as assets to cities. In this paper, a revenue oriented, cloud microservice based air quality prediction framework is proposed for smart cities. This framework collects the air quality parameters such as SO2, NO2, RSPM, and so forth, for different cities in a mongodb based cloud database; it develops air quality prediction models based on Random Forests (RF); and it predicts the air quality parameter values accordingly. For each access to the proposed air quality prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city stations are calculated based on the prediction queries. In the experiments, the prediction framework for predicting SO2, NO2, and RSPM attained over 70 to 90 percentage of prediction accuracies. Keywords-Air quality; Revenue model; Random forests; Pre- diction algorithm; I. I NTRODUCTION Establishing smart cities has become a critical notion for major city authorities in developing or developed countries. This is due to the fact that the population ratio of urban to rural locations has dramatically increased in recent years. These authorities are, therefore, motivated to improve the economic situations and control the over-utilization of city assets, including water, electricity, and so forth. To do so, the utilization of various modern ICT technologies, including cloud or edge or mobile technologies are suggested by various city architects/planners. One of the major challenges endeavored in modern cities of India or China is to control air pollutions. In fact, the cause for air pollutions has its roots in wrong industrial practices. Notably, WHO reports that around 92 percentage of the global population lives in the air polluted locations which lead to millions of deaths – SO2 could lead to a lung cancer. Towards this end, several inter-disciplinary research efforts have led to a few production level monitoring schemes have been developed in recent years in order to assist city residents or authorities. For instances, an air quality forecast software of Siemens periodically informs users (in advance) about the pollution level of a city [10]; similarly, the World Air Quality Indexing team has developed a social initiative project for providing the visual representation of the air quality information to the city residents [1]. However, in reality, there is a lack of a business model in the existing approaches. Smart city authorities, especially from developing countries, should reap financial benefits for actively initiating such eco-concerned societal projects. In addition, due to the advancement of sensor technologies, air quality sensors have become inexpensive in recent years. Accordingly, several air quality monitoring stations have been established worldwide. For instance, EPA from USA has established over 10000 air quality monitoring stations. Consequently, the establishment of these air quality monitoring stations has become a source of challenges while i) handling multiple user queries, ii) storing sensor data in a distributed fashion, and iii) diligently predicting air quality information from different sites. Undoubtedly, modern ICT technologies could address the above mentioned challenging scenarios. This paper proposes an innovative revenue-oriented air quality prediction framework for smart cities. In this proposed approach, cities are credited for each air quality status query received from global Internet users (which may be associated with a bitcoin approach in the future). The city authorities could in turn formulate protective schemes or control measures or penalize industries responsible for the same. The proposed approach, therefore, creates revenue to smart cities and con- trols the pollution level of cities. The air quality prediction framework, proposed in this paper, is assisted with Random Forest (RF) algorithm based cloud microservices (which model air quality parameters such as Sulphur dioxide (SO2), Suspended Particulate Matter (SPM), Respirable Suspended Particulate Matter (RSPM), Carbon Monoxide (CO), and so forth, from various monitoring stations). Later, utilizing the trained model, the air quality parameter value is predicted for any testing queries. The proposed prediction framework is exposed as REST based cloud services. The conducted experiments manifested the strength of random forest prediction algorithm in the context of the air quality predictions. In succinct, the contributions of the paper are listed as follows: 1) Revenue based air quality prediction framework that creates credits to smart cities is proposed using cloud
Transcript
Page 1: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

Revenue Oriented Air Quality PredictionMicroServices for Smart Cities

Shajulin BenedictIndian Institute of Information Technology Kottayam,

(Earlier - Guest Professor, TUM – Germany),Email: [email protected]

Abstract—A notion of creating smart cities with integratedhigh end ICT technologies, including cloud, edge, IoTs, and soforth, has increased recently in order to provide a reasonablyhigh comfort to city residents with limited human intervention.In fact, one of the major challenges faced by city authoritiesof developing countries is to control air pollution, which mightcause chronic diseases and financial concerns to the residents.Although a few air quality monitoring and air pollution controlmeasures exist, these solutions are not remaining as assets tocities. In this paper, a revenue oriented, cloud microservice basedair quality prediction framework is proposed for smart cities.This framework collects the air quality parameters such as SO2,NO2, RSPM, and so forth, for different cities in a mongodb basedcloud database; it develops air quality prediction models based onRandom Forests (RF); and it predicts the air quality parametervalues accordingly. For each access to the proposed air qualityprediction framework, revenue oriented cloud microservices,written in golang, are initiated and the credits for city stations arecalculated based on the prediction queries. In the experiments,the prediction framework for predicting SO2, NO2, and RSPMattained over 70 to 90 percentage of prediction accuracies.

Keywords-Air quality; Revenue model; Random forests; Pre-diction algorithm;

I. INTRODUCTION

Establishing smart cities has become a critical notion formajor city authorities in developing or developed countries.This is due to the fact that the population ratio of urbanto rural locations has dramatically increased in recent years.These authorities are, therefore, motivated to improve theeconomic situations and control the over-utilization of cityassets, including water, electricity, and so forth. To do so,the utilization of various modern ICT technologies, includingcloud or edge or mobile technologies are suggested by variouscity architects/planners.

One of the major challenges endeavored in modern cities ofIndia or China is to control air pollutions. In fact, the causefor air pollutions has its roots in wrong industrial practices.Notably, WHO reports that around 92 percentage of the globalpopulation lives in the air polluted locations which lead tomillions of deaths – SO2 could lead to a lung cancer. Towardsthis end, several inter-disciplinary research efforts have led to afew production level monitoring schemes have been developedin recent years in order to assist city residents or authorities.For instances, an air quality forecast software of Siemensperiodically informs users (in advance) about the pollution

level of a city [10]; similarly, the World Air Quality Indexingteam has developed a social initiative project for providing thevisual representation of the air quality information to the cityresidents [1].

However, in reality, there is a lack of a business model inthe existing approaches. Smart city authorities, especially fromdeveloping countries, should reap financial benefits for activelyinitiating such eco-concerned societal projects. In addition,due to the advancement of sensor technologies, air qualitysensors have become inexpensive in recent years. Accordingly,several air quality monitoring stations have been establishedworldwide. For instance, EPA from USA has establishedover 10000 air quality monitoring stations. Consequently, theestablishment of these air quality monitoring stations hasbecome a source of challenges while i) handling multiple userqueries, ii) storing sensor data in a distributed fashion, andiii) diligently predicting air quality information from differentsites. Undoubtedly, modern ICT technologies could addressthe above mentioned challenging scenarios.

This paper proposes an innovative revenue-oriented airquality prediction framework for smart cities. In this proposedapproach, cities are credited for each air quality status queryreceived from global Internet users (which may be associatedwith a bitcoin approach in the future). The city authoritiescould in turn formulate protective schemes or control measuresor penalize industries responsible for the same. The proposedapproach, therefore, creates revenue to smart cities and con-trols the pollution level of cities.

The air quality prediction framework, proposed in thispaper, is assisted with Random Forest (RF) algorithm basedcloud microservices (which model air quality parameterssuch as Sulphur dioxide (SO2), Suspended Particulate Matter(SPM), Respirable Suspended Particulate Matter (RSPM),Carbon Monoxide (CO), and so forth, from various monitoringstations). Later, utilizing the trained model, the air qualityparameter value is predicted for any testing queries. Theproposed prediction framework is exposed as REST basedcloud services. The conducted experiments manifested thestrength of random forest prediction algorithm in the contextof the air quality predictions. In succinct, the contributions ofthe paper are listed as follows:

1) Revenue based air quality prediction framework thatcreates credits to smart cities is proposed using cloud

Page 2: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

microservices.2) The efficiencies of the proposed random forest algorithm

based air quality prediction services were experimentedwith the available air quality measurement datasets. Thefocus of this work, therefore, relied on the application ofmodern ICT technologies for predicting the air qualityparameters.

The rest of the paper is explained as follows: Section IIdiscusses the related works; Section III explains the proposedrevenue based air quality prediction framework; Section IVvalidates the random forest based air quality prediction algo-rithm with experimental results; and, Section V lists a fewconclusions and outlooks of this work.

II. RELATED WORKS

In general, discussions about smart cities were inspired bya few economists in the past [5]. However, the concept ofsmart cities has taken a serious momentum in recent yearsowing to several initiatives/communities such as intelligentcommunities. In addition, several city ranking schemes ( [4],[8]) motivated city authorities to put forth necessary effortsfor attaining top grades. In fact, introducing the city rankingschemes has impacted the way of thinking cities in variouspositive dimensions: i) improving broadband networks, ii)carefully utilizing space while accommodating more habitants,iii) providing intelligent alarm / alerts, iv) distributing suf-ficient low cost security systems, v) improving policies, vi)increasing the soft skills, vii) reducing air/water pollutions,and so forth.

Generally, air quality pollutions are categorized in termsof outdoor, indoor, and industrial air quality indexes. Severalresearch efforts were carried out in the past for analyzing theair quality parameters while targeting the health conditions ofhumans or plants. Notably, Kadri et al [6] proposed a machine-2-machine approach of monitoring air quality parameters in adistributed fashion. Recently, Yash et al [11] have proposed acloud assisted air quality monitoring approach.

In terms of air quality predictions, a few researchers havefocused on predicting the air quality parameters such as NO2,SO2, RSPM, ozone depletion, and so forth, using predictionalgorithms such as SVM, EPLS, and neural networks. Forinstances, authors of [9] have studied the application of neuralnetwork based prediction algorithm for predicting air qualityparameters; Pietro et al [7] have analyzed the impact of theseverity of CO and NO2 along roadsides using the neuralnetwork prediction algorithm; authors of [12] have exploredthe effect of utilizing EPLS method for clustering air qualitydata. In addition, authors of [3], have created pure analyticalmodels for analyzing the air quality of a specific region. In thiswork, a revenue oriented cloud microservice based air qualityprediction framework is proposed.

III. REVENUE BASED AIR QUALITY PREDICTIONFRAMEWORK

This section describes the proposed air quality predictionframework which includes cloud microservices, random forest

modeling/prediction algorithm and REST based service imple-mentation in detail (Figure 1).

A. Framework

The proposed framework consists of the following entities:i) air quality monitoring sites, ii) sensor gateways, iii) cloudmicroservice units, iv) cloud database, v) public cloud con-nectivity.

1) Air Quality Monitoring: Air quality parameters such asSulphur dioxide (SO2), Suspended Particulate Matter (SPM),Respirable Suspended Particulate Matter (RSPM), CarbonMonoxide (CO), and so forth, should be monitored in adistributed fashion from monitoring sites using a few airquality monitoring gas sensors such as CO2 analyzer and SO2analyzer. The sensed data from these sensors are propagatedusing wireless communication protocols, namely, 3G, 4G, orZigbee protocols to the nearest sensor gateways.

2) Sensor Gateways: The sensor gateways are, in general,computationally powerful than sensors. These gateways are re-sponsible to channelize the user query requests to appropriatesensor units. In addition, the gateways route the sensed datato the cloud server units for future references.

3) Cloud microservice units: This work provides moreemphasis on the proposed cloud microservices which arehosted on cloud servers (either public cloud server units fromAmazon EC2 or Google Compute Engines or a privatelyowned cloud server units). The microservice units are detailedlater in the section.

4) Cloud database: The sensed data are stored in a clouddatabase which was installed using a mongodb. The databasein the framework could be configured to handle multipleservice queries in a distributed fashion.

5) Public cloud connectivity: This entity is responsible forrequesting additional computational from public clouds suchas, Amazon EC2. This provision is meant for the scalabilitypurpose of the framework. The scalability feature is importantwhen billions of users request for the air quality predictionservice, especially in events such as Diwali or newyear events(when pollution happens due to fireworks).

B. Cloud Microservices

In general, a cloud microservice is the smallest possiblecomplete service, which is usually hosted in a cloud VM. Itlistens to and serves client requests. The cloud microserviceapproach is introduced in the proposed framework so thatservices could be updated or modified at any time with uninter-rupted downtime. This paper proposed four microservices: a)microservice frontend, b) querylog service, c) revenue service,and d) RF prediction service. These services are hosted oncloud servers.

1) Microservice frontend: This service remains as a inter-face portal to users. The following activities are carried outby this service: i) It invokes time to time measurements fromvarious air quality monitoring stations and stores the valuesin database. However, in this study, the priorly measured airquality parameter values from various available lists of the

Page 3: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

Fig. 1. Revenue Service Oriented Air Quality Prediction Framework

monitoring stations of India are stored in the database insteadof directly measuring them. ii) The frontend is responsible forinvoking the other services such as querylog service, revenueservice, and prediction services.

2) Querylog service: Querylog service records the nameof the city stations that are requested by users for analyzingthe air quality stations. For instance, if a user requests for theair quality status of Adayar-Chennai station, the name of thestation will be recorded in the cloud database. In addition, thenumber of query Nqi for this station i will be incremented.

3) Revenue service: The revenue service collects the num-ber of queries received for each city agencies from thedatabase. Later, it prepares a table with the credits for eachstations. The revenue is calculated as follows:

Ri = 50 ∗Npi ∗Nqi ⇐⇒ Nqi < 50

Ri = 63.3 ∗Npi ∗Nqi ⇐⇒ Nqi >= 50(1)

where, Ri is revenue credited for city station i; Npi is thenumber of measured air quality parameters; and, Nqi is thenumber of received queries (the testing dataset).

4) RF-Prediction service: Immediately when a user re-quests for checking the air quality of a city, the service requestis handled to the RF-Prediction service. The RF predictionservice collects previous air quality measurement data fromthe cloud database and models the importance of air qualityparameters and their corresponding sensed values based onthe random forest (RF) modeling algorithm. Later, the servicecollects the RF modeled values from cloud database andpredicts the air quality parameter value. The RF predic-tion services are further branched into three sub-predictionmicroservices, namely, RF-SO2-prediction service, RF-NO2-prediction service, and RF-RSPM-prediction service. Theseprediction microservices are implemented using the golanglanguage of Google.

C. RF Modeling and Prediction Algorithm

This subsection describes the Random Forest algorithm indetail. Random Forest modeling and prediction algorithmspredict air quality parameter values. In general, RandomForest algorithm is a non-parametric decision tree based al-gorithm [2].

1) RF Modeling algorithm: Random Forest modeling al-gorithm is utilized to construct trees, the graph based nodes,based on an ensemble supervised learning approach. The algo-rithm creates trees during the bagging process such that the airquality parameter values (independent variables) are dispersedwith high variances. On bagging process, RF algorithm picksup a few samples (s1, s2, ..., sn, where s1−n ∈ S) from thecloud database and identifies the appropriate trees based onthe independent air quality parameter variables (apvn). Thedependent variable is any one of the air quality parametervalue – RF-SO2 or RF-NO2 or RF-RSPM; the independentvariables are the air quality parameter values that are notdependent variables for the modeling/prediction. At the end ofthis algorithm execution, a learning model will be developedwhich is represented as RFt. In addition, a few newer samplessm are validated with RFt.

2) RF Prediction algorithm: RF prediction algorithm pre-dicts the air quality parameter value based on the modelgenerated during the RF modeling service. During the predic-tion phase, Random forest algorithm creates regression pointsbased on the independent variables of the test set (the queryreceived from the cloud user); it examines which tree closelyrelates to the incoming air quality parameter values apvi .And, finally, the error rate of the predictions is calculatedusing the mean squared error R2 for the independent variableestimations.

Page 4: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

D. Micro Services

As mentioned earlier, in this work, the RF predictionalgorithms and revenue approaches are implemented as cloudmicroservices for the proposed air quality prediction frame-work. Three microservices namely RF-SO2 prediction service,RF-NO2 prediction service, and RF-RSPM prediction serviceare developed as part of the prediction services.

Due to this microservice based implementation style, mul-tiple standalone modeling and prediction services could listenindependently for user requests.

1) REST Golang Implementation: Each of the micro-services specified in Figure 1, including the prediction ser-vices, is exposed to the external world using RepresentationalState Trasfer protocol (REST) golang-based service imple-mentation. The services communicate each other using httpprotocols based on the web service verbs such as PUT, POST,and GET operations.

2) Revenue Microservice: The revenue microservice doesthe following activities for every five minutes: i) it connects tothe cloud mongodb database, ii) collects the number of queriesreceived for each city stations, iii) calculates the credits foreach city stations, and iv) reports on the calculated credits tothe city authorities.

IV. EXPERIMENTAL RESULTS

The following section explores i) the experimental setup,ii) the revenue generated for each city stations, and iii) theimpact of RF predictions.

A. Experimental Setup

In this work, the microservices, specified in Figure 1,including the revenue service, were hosted on cloud VMs.These services were invoked using service handlers imple-mented in golang. The service handlers of the predictionservices got connected to the R based implementation of RFalgorithm, which was fed by the data from cloud databaseof the framework. In the experiments, the Mongodb basedcloud database was installed in another VM which had adedicated IP address throughout the experiments. The reasonfor choosing the cloud VMs was to have a scalable solutionfor the proposed software framework – for instance, scalabilitycould be achieved with the invocation of configurable VMsusing kubernetes clusters. In fact, these microservices couldalso have been hosted on independent VMs or on independentphysical servers. However, choosing dedicated physical serversmight not be a cost effective solution.

In order to manifest the proposed revenue oriented air qual-ity prediction approach, the pre-measured air quality parametervalues recorded in 2015 for four southern states of India,namely, Karnataka, Kerala, Tamilnadu, and AndraPradesh 1

were considered.

1https://data.gov.in/catalog/historical-daily-ambient-air-quality-data

Fig. 2. SO2 Prediction Results using RF Prediction Service

B. Air Quality Prediction and Revenue

In order to validate the efficiency of the proposed RandomForest prediction service based mechanism, at first, 50 per-centage of the air quality measurement data that are availablefor the southern states of India namely Tamilnadu, Kerala,Karnataka, and AndraPradesh were loaded into the databasein an equi-distant sampling fashion. To assume the real casesituations, the Microservice frontend had a boot mechanism toautomatically submit the air quality prediction query requestto the database. These processes lead to 3960 entries in thedatabase. Thus, the measurement data, namely, SO2, NO2,RSPM, and RSPM/PM10 for different cities and the corre-sponding station code were recorded as training dataset in thedatabase.

For each query requests, the QueryLog service recordedthe corresponding city station name and the contact details(test emails) of the city authorities to the cloud databasewhich was later utilized by the revenue service to notify them.Accordingly, the revenue service calculated the credits for citystations. Table I shows the first few entries of city stations andtheir corresponding credits which were calculated based on theequation 1 during the experiment.

The prediction results for the queries are depicted in Fig-ures 2, 3, and 4. The x-axis is the numeric conversion ofair quality measurement date and space – i.e., the date wasconverted to numeric values and it was summed with the citystation code; and, the y-axis corresponds to the air quality pa-rameter value. In the Figures, the points represent the observedvalues and the lines are the predicted values of the air qualityparameters. In addition, the prediction service calculated theerror rate of the RF predictions. It was observed that the RF-SO2 prediction service, RF-NO2 prediction service, and RF-RSPM prediction services achieved R2 as follows: 90.14, 71.9and 70.26.

Page 5: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

TABLE IREVENUE CREDITED TO CITY STATIONS

Sl.No. City Station Number of queries Credits in Rs.1 Compound of APSPCB office building 11 16502 Compound of PCCF’s Office building 8 12003 B.Ed College (PKCE), Nehru Nagar, Karaikal 53 100704 Chamber Of Commerce, Pondicherry 35 52505 DSTC Office Upstairs, PHB 3rd Floor, AnnaNagar. Pondicherry-5. 34 51006 Govt. Tourist Home, Kovilpathu, Karaikal 43 64507 M/s Puducherry Power Corporation Limited, Polagam, T.R. Pattinam, Karaikal 46 69008 PIPDIC Ind. Estate Mettupalayam, Pondicherry 34 51009 Adyar, Chennai 42 630010 Anna Nagar, Chennai 44 660011 AVM Jewellery Building, Tuticorin 38 570012 Bishop Heber College, Tirchy 24 360013 Central Bus Stand, Trichy 16 240014 District Environmental Engineer Office, Imperial Road, Cuddalore 43 645015 Distt. Collector’s Office, Coimbatore 47 705016 Eachangadu Villagae 40 600017 Fenner (I) Ltd. Employees Assiciation Building Kochadai, Madurai 42 630018 Fisheries College, Tuticorin 32 480019 Gandhi Market, Trichy 23 345020 Golden Rock, Trichy 20 300021 Govt. High School, Manali, Chennai. 43 645022 Highway (Project -I) Building, Madurai 39 585023 Kathivakkam, Municipal Kalyana Mandapam, Chennai 40 600024 Kilpauk, Chennai 40 600025 Kunnathur Chatram East Avani Mollai Street, Madurai 36 540026 Madras Medical College, Chennai 43 645027 Main Guard Gate, Tirchy 22 330028 NEERI, CSIR Campus Chennai 45 675029 Poniarajapuram, On the top of DEL, Coimbatore 35 525030 Raja Agencies, Tuticorin 32 480031 Raman Nagar, Mettur 50 950032 SIDCO Industrial Complex, Mettur 45 675033 SIDCO Office, Coimbatore 37 555034 SIPCOT Industrial Complex, Cuddalore 33 495035 Sowdeswari College Building, Salem 60 1140036 Thiruvottiyur Municipal Office, Chennai 33 495037 Thiruvottiyur, Chennai 43 645038 Thiyagaraya Nagar, Chennai 37 555039 Vallalar Nagar, Chennai 22 3300Contd... ... ... ...

Fig. 3. NO2 Prediction Results using RF Prediction Service Fig. 4. RSPM Prediction Results using RF Prediction Service

Page 6: Revenue Oriented Air Quality Prediction MicroServices for ... · prediction framework, revenue oriented cloud microservices, written in golang, are initiated and the credits for city

V. CONCLUSION

Formation of smart cities has evolved as an interdisciplinaryresearch as these cities require ICT technologies, environmen-tal research, and economics for efficiently architecting them.Although a few air quality monitoring approaches exist in liter-ature, they are either not utilizing modern ICT technologies ornot profitable solutions. This paper proposed revenue-orientedcloud microservice based air quality prediction frameworkusing random forest algorithms. The proposed prediction ap-proach was experimented using the air quality monitoring dataof four southern states of India. The microservices predictedwith an accuracy of over 70 to 90 percentage of R2 values.

REFERENCES

[1] Air Quality Indexing software, https://waqi.info/, 2017.[2] Leo Breiman, Random Forests, in Machine Learning, Vol. 45, No.1, pp.

5-32, 2001.[3] Cai, C.-J., X. Zhang, K. Wang, Y. Zhang, L.-T. Wang, Q. Zhang, F.-K.

Duan, K.-B. He, and S.-C. Yu, Incorporation of New Particle Formationand Early Growth treatments into WRF/Chem: Model Improvement,Evaluation, and Impacts of Anthropogenic Aerosols over East Asia,Atmospheric Environment, Vol.124, pp.262-284, 2016.

[4] Giffinger R., Fertner C., Kramar H., Kalasek R., Pichler-Milanovi N.,Meijers E., Smart cities: Ranking of European medium-sized cities, http://www.smart-cities.eu/download/smart cities final report.pdf, 2007.

[5] Jessie M. Shapiro, Smart Cities: Quality of Life, Productivity, and theGrowth Effects of Human Capital, in the Review of Economics andStatistics, Vol. 88, No. 2, pp- 324-335, 2006.

[6] A. Kadri, E. Yaacoub, M. Mushtaha, and A. Abu-Dayya, Wireless sensornetwork for real-time air pollution monitoring, in Proc. of IEEE Int.Conf. on Commn., Signal Proc. and their Appln, pp. 1 - 5, 2013.

[7] Pietro Zito, Haibo Chen, and Margaret C. Bell, Predicting Real-TimeRoadside CO and NO2 Concentrations Using Neural Networks, inIEEE Transactions on Intelligent Transportation Systems, Vol. 9, No.3, pp.514-522, 2008.

[8] Rolland Busch, The Green City Index, in www.siemens.com/greencityindex, 2017.

[9] Santosh Kumar Nanda, Debi Prasad Tripathi, S.S. Mahapatra, Ap-plication of Legendre Neural Network for Air- Quality Prediction,International Conference on Engineering and Technology, May 2011.

[10] Siemens Air Quality Forecasts, https://www.siemens.com/innovation/en/home/pictures-of-the-future/infrastructure-and-finance/smart-cities-air-pollution-forecasting-models.html, 2017.

[11] Yash Mehta, Manohara Pai M.M., Sanoop Mallissery, ShwetanshuSingh, Cloud enabled Air Quality Detection, Analysis and Prediction- A Smart City Application for Smart Health, in Proc. of 3rd MECInt.Conf. on Big Data and Smart City, pp. 1-7, 2016.

[12] Yunliang Chena f., Lizhe Wanga, Fangyuan Lia, Bo Dua, Kim-KwangRaymond Choob, Houcine Hassand, Wenjian Qine, Air quality dataclustering using EPLS method, Vol. 36, pp. 225232, 2017.


Recommended