Traffic Pattern Analysis from GPS Data: A Case Study of ... Pattern... · per square kilometer in the heart of the city [1]. Dhaka city plays an important role in Bangladesh’s

Traffic Pattern Analysis from GPS Data: A CaseStudy of Dhaka City

Md Maksudur Rahman∗, M M Mahmudunnabi Shuvo†, Moinul Islam Zaber‡ and Amin Ahsan Ali§Department of Computer Science and Engineering, University of Dhaka.

Dhaka, Bangladesh.Email:{∗2012-71-2014, †2012-91-2030}@students.cse.univdhaka.edu, {‡zaber, §aminali}@du.ac.bd

Abstract—Traffic congestion is one of the most alarmingproblems of Dhaka, the capital of Bangladesh. However, notmuch work had been done on traffic pattern modeling for Dhakacity. In this paper, we analyze traffic intensity pattern computedfrom GPS data. The data contains traffic intensity informationfor 11,769 road segments over 15 days. We analyze the impact ofmarketplaces, number of road intersections, and having rickshawfree roads on the traffic intensity. In order to analyze the trafficpattern at a macroscopic level, we analyze the traffic patternof the 13 zones of the city proposed by RAJUK, the authorityresponsible for the development of Dhaka. For each zone weinvestigate the impact of a number of different factors, e.g., landuse, number of bus routes, number of road intersections, on thetraffic intensity.

Keywords—Traffic data analysis; Land use pattern; GlobalPositioning System (GPS); Traffic congestion; Zone cluster.

I. INTRODUCTION

About 17 million people live in the greater Dhaka in anarea of 1,528 square kilometers with more than 45,000 peopleper square kilometer in the heart of the city [1]. Dhakacity plays an important role in Bangladesh’s economy. But,traffic congestion is an alarming problem that hampers theeconomic growth of Bangladesh. The financial loss from trafficcongestion in Dhaka city is well over $18.5 million per day[2]. Only 8% of Dhaka’s total land area is being used as roadnetwork where smooth traffic system demands 25% of thecity’s surface area [3]. People spend on an average 2.35 hoursin traffic activities daily, of which more than a half of that(1.30 hours) is eaten up due to traffic congestion [4]. It isestimated that in 2016, the average traffic speed in Dhakais 6.4 kilometers per hour but without proper planning andsubstantial public transport investment, the average speed mayfall to 4.7 kilometers per hour by 2035 [5]. Although 37kilometers long railroads pass through Dhaka city, it has verylittle contribution to the city’s transport system [6].

In order to address such appalling conditions, in 2009-10 a preparatory survey on Dhaka city for urban transportnetwork development Bangladesh was conducted by JapanInternational Cooperation Agency (JICA) [7]. The results fromthe survey provides an overview on the future planning oftransport network of Dhaka. This study also analyzed Dhakacity’s current traffic condition. This was a observational fieldstudy and interview based survey and therefore, is proneto human error specially in the observational field study.Conducting such a study is also expensive and thus cannot

be carried out regularly to monitor the traffic patterns of acity over a long period.

A feasible alternative to such surveys can be the use ofGPS traces from vehicles to obtain real time traffic data. Inthis paper, we have used the GPS (Global Positioning System)record obtained from taxicabs running through Dhaka city. Tothe best of our knowledge, traffic modeling using such data hasnot been carried out in Bangladesh. The data has been providedby Gobd 1. The traffic intensity information computed fromthe speed and altitude of the gps data and road class (roadwidth proxy) obtained from OSM (Open Street Map) whichhas the road class data in their map data. The intensity values,associated with each road segment are in the range of 0 to1. Each datapoint also contains a timestamp. We preprocessthis intensity data (e.g., removing outliers) and use it to findmacroscopic traffic patterns.

Traffic pattern modeling and finding the factors that influ-ence traffic congestion are important issues in urban road net-work development. In recent years, Macroscopic FundamentalDiagrams (MFD) have been used to model traffic flow incities such as Brisbane, Australia [8]. However, in order todo such modeling, information on the number of vehicles thatrun through a road intersections at a certain time is required.Such dataset however is not available for Dhaka. Therefore, forinvestigating traffic pattern at a macroscopic level, we makeuse of land use and social infrastructure data that has directeffect on travel demand [9]. We use clustering method to findthe similar zones (using land use pattern) of Dhaka city andthe characteristics of traffic intensity of each cluster. We makeuse of Google maps and QGIS software to select areas foranalyzing the effect of social infrastructures as market placeson traffic congestion. Road segment intersections (determinedfrom latitude, longitude of road segment) are also considered.Similar to studies on trasportation systems in cities in India[10], we collected the bus routes and the rickshaw free routesof Dhaka city for analyzing the influence of transportationmode on traffic congestion. The four main contributions ofthis paper can be summarized as follows.

• Finding traffic intensity pattern over hours of the day anddays of the week.

• Finding zone clusters and analyze traffic pattern analysisfor each cluster.

1Gobd.co provides real time traffic information with route suggestions.

• Identifying factors (e.g., land use, number of road inter-sections, bus routes, and social infrastructures) that maycause traffic congestion.

• Regression analysis for finding the impact of factors ontraffic congestion.

In section II, we describe the related work on traffic model-ing. Data collection and preprocessing steps to clean the dataare described in section III. Section IV describes the trafficpattern analysis using the GPS data. We conclude our paperin section V.

II. RELATED WORK

A number of works have been done on traffic modeling thataims to find ways to reduce traffic congestion.

Geroliminis et al. [11] have proposed an observation basedmodel. They described macroscopic relations between trafficvariables for a single link and also presented MFD for largeroad network. However, real time traffic observations (numberof vehicles plying on the road) are needed to build themacroscopic traffic modeling that they have proposed.

In absense of such data, GPS-equipped vehicles can beused as traffic sensors. Large-scale GPS traces can be usedto estimate traffic congestion and the underlying dynamics ofroad network. Liu et al. [12] described intra urban land usevariations using traffic patterns. The temporal variations ofboth pick-ups and drop-offs were determined. They analyzedthe GPS data of more than 6600 taxis. Vacca et al. [13]analyzed the route switch behavior of vehicles. The paperdescribes the main attributes of the routes and the character-istics of the users that most influence the choice of multipleroutes for the same origin-destination trip. This work makesuse of 361 morning commute trips GPS data collected in themetropolitan area of Cagliari (Italy) by taxicabs. Castro el al.[14] discussed a predictive model. The authors also used GPSdata from vehicles to predict future traffic conditions. Yoonet al. [15] estimated traffic congestion from GPS data usingtemporal and spatial information. However, none of the worksdescribe traffic pattern over a macroscopic region.

The Dhaka urban transport network development study(DHUTS) [7] does, on the other hand, collect and analyzetraffic data over the whole city. In this study, a HouseholdInterview Survey was conducted to collect the data of travelbehavior patterns throughout the DMA (Dhaka MetropolitanArea). The survey provides the transportation mode choicewith respect to different income level. The study also providesthe analysis on resident movement across the DMA, DCC(Dhaka City Corporation), and RAJUK (Rajdhani UnnayanKartripakkha) areas. The study shows that the Trip ProductionRate defined as trips/day is around 2.74 in Dhaka city. Thedraft of Dhaka Structure Plan (2016-2035) [16] publishedby RAJUK also provides an overview of daily passengermovement in Dhaka. It also publishes some policies in orderto control traffic congestion. However, none of these studiesprovide models describing the traffic patterns over time orinformation on the factors that influence it. Hanson et al. [17]

described an overview of the relationships between transporta-tion and land use and examine the impact of land use on thetransportation. They also provide a brief description of factorsto consider when examining land use pattern and the impactsof it on transportation investments. The land use pattern forBangladesh can be obtained in the final report of RAJUK [18],the fifth of the series of the reports submitted under the DAP(Detailed Area Plan) which covered 108.97 sq.km of Dhaka(covering most of the city). The entire area is divided into 13DPZ (Detailed Planning Zone) zones. The area was dividedconsidering population density and homogeneity of the landuse. We use the land use pattern and social infrastructuresinformation from this report for macroscopic traffic patternanalysis.

III. DATA COLLECTION AND PREPROCESSING

In this section, we describe the data set and the prepro-cessing steps. First, we divide data for each road segmentinto 30 minutes windows and remove outliers. We then createthe required shape files to plot the data on map and retainthe information of different zones. In order to perform macrolevel analysis at the zone level, we find the road segmentsinformation of each zone. We also obtain the number ofrickshaw free roads and the number of bus routes in Dhakacity. GPS data usually contains latitude, longitude, speed,altitude and time-stamps information. However, the data set weobtained contains traffic intensity for different road segmentsof Dhaka city.

A. Data description

The data was provided by a private company named Gobd 1

which is currently working on a route suggestion application.The data we have, has the traffic intensity information from 1stSeptember, 2015 to 15th September, 2015. The traffic intensityis measured based on the road class (road width proxy), speedand altitude of the GPS data using OSM (Open Street Maphas the road class data in their map data). A single road isdivided into multiple road segments and there are 11,769 roadsegments in total. There are 16,23,280 records in whole data.Data descriptions and data types are give in table I.

TABLE I: Data descriptions and types.

Field Description Data typeId Identifier of the log record. ObjectTimeStamp Time stamp when the record

was created.Date - Time

OsmName Human readable name for theroad segment.

String

Intensity The intensity of traffic jam onthe road segment within 0 to 1.

Float

Path Starting and ending coordinatesof the road segment.

Lat/Long range

DistanceKM The length of the road segmentin Kilometers.

Float

Cost The utility of the road segmentas travel route ((range 0 to 1).

Float

RoadSpeedKMH Expected speed of road seg-ment at Kilometers per Hour.

Interger

B. Data Preprocessing

For traffic analysis, we perform a number of preprocessingsteps. First, we remove the road segment records containinginsufficient data. In order to do this, we divide the 24 hours ofa day into 48 windows each having a length of 30 minutes. Weonly use data for further analysis if a road segment has at least5 records in the 30 minutes windows in 15 days. After cleaningthe data, we have 8,966 road segments. Second, we find andremove the outliers. We compute arithmetic mean and standarddeviation of traffic intensity for 30 minutes time windows fromthe data we have for each road segment. We consider a recordas outlier if the traffic intensity is not within the 2-standarddeviation from the arithmetic mean for that road segment inthat particular time window (30 minutes). Although outliersmay represent extreme traffic congestion in a time window,we opted to find the general traffic pattern. Finally, in order toplot data on map, we use QGIS and Google Map. We generatea csv file containing all the latitude, longitude information ofroad segments from our data set. Then we convert it to kmlfile using Google Earth’s earthpoint. We draw a polygon onGoogle Map and then convert it as kml file from map options.Then we load both of our kml files in QGIS as vector layerand export them as a shape file.

Fig. 1: Road segments and study area (green colored area)after removing outliers.

Figure 1 shows the road segments after removing outliersfrom the data set.

C. Zoning of Dhaka City

From RAJUK report [18], the Dhaka city is divided into13 DPZ (Detailed Planning Zone) zones. The land use infor-mation is also given for each zone. Most of the DCC (DhakaCity Corporation) wards are covered in this zoning scheme.DPZ-1 is the western part of old Dhaka and DPZ-2 is theeastern part. DPZ-3, DPZ-4, DPZ-6 and DPZ-7 are mainlycommercial areas. DPZ-5 situated at the eastern fringe. DPZ-8, DPZ-9, DPZ-10 are the suburb areas. DPZ-11, DPZ-12,DPZ-13 are mainly residential areas.

Distributing Road Segments among Zones: We analyze ifa road segment is in a particular zone or it is a connecting path

between zones. From our data set, we have latitude, longitudeinformation of all road segments. We use this informationto distribute the road segments among zones. In order to dothis, if all the coordinates of a particular road segment arelocated in a zone’s area then we consider that particular roadsegment as a property of that zone. If the coordinates of a roadsegment are located in more than one zone then we considerthat road segment as a connection path between those zones.We take each zone as polygon and check if all the points of aroad segment are in one polygon or distributed among severalpolygons.

Rickshaw Free Roads and Bus Routes of DPZ Zones:We find the number of bus routes for each DPZ zone. We usethe information collected from BRTA2 official website [19].There are 268 bus routes in total. JICA report [7] has theinformation of the rickshaw free roads of Dhaka city. Weuse that information finding the latitude, longitude in orderto determine the rickshaw free road segments from our data.There are 22 rickshaw free road segments in our data set.

IV. TRAFFIC ANALYSIS

In this section, we describe different traffic pattern analysis.First, we show traffic intensity pattern for different cases(whole day traffic pattern, day wise traffic intensity variation,effect of marketplaces, rickshaw free roads traffic intensitypattern). Road segment intersections are determined to showthe effect of road intersections on traffic intensity. Then,clustering methods are used to find the similarities amongzones. Finally, we perform regression analysis using differentfactors that might affect traffic intensity.

A. Overview of Dhaka’s Traffic Intensity Pattern

We find that the traffic pattern differs with respect to timein a day. We take the average of traffic intensity for all timewindows for that experimental 15 days. Figure 2 shows theintensity average of all road segments at different times of theday.

Fig. 2: Overview of traffic intensity (overall Dhaka city). Itdisplays the traffic intensity (average) pattern (blue curve) for24 hours of Dhaka city. The traffic intensity is high in workinghours rather than late night.

From figure 2, we observe that Dhaka suffers from heavytraffic intensity in most of the time of a day. After 6 pm, the

2Bangladesh Road Transport Authority (BRTA) is a regulatory body tocontrol manage and ensure discipline in the road transport sector.

traffic intensity becomes low. This is a general traffic patternof Dhaka city where we consider average traffic intensity overall the road segments. We further investigate the pattern atdifferent days and zones to find how the pattern changes.

B. Comparison between Weekdays and Weekends Traffic In-tensity

Here, we investigate how the traffic intensity pattern changesat different days of a week. In Bangladesh, the governmentoffices are kept closed at Friday and Saturday. So, we con-sider Fridays and Saturdays as weekends. From the day wiseintensity graph, we see that traffic intensity is low at weekends.

Fig. 3: Traffic pattern for weekdays and weekends.(a) Averagetraffic intensity pattern for weekdays (blue dashed curve) andweekends (red solid curve). (b) Average traffic intensity fordifferent days of a week. Traffic intensity is lower at weekendsthan weekdays.

Figure 3 (a) shows us that the traffic intensity is high atweekdays (blue dashed curve) than weekends (red solid curve).Figure 3 (b) illustrates the average traffic intensity of each dayin a week. It is clear that traffic intensity is much lower atweekends (Fridays and Saturdays).

C. Effect of Marketplaces on Traffic Intensity

There are many marketplaces beside the roads of Dhakacity. We select a central area of Dhaka city in order to findthe effect of marketplaces on traffic congestion. We select thatparticular study area because it contains a large number ofshopping centers. Figure 4 shows our study area for analyzingmarketplaces effect on traffic intensity. We select the roadsegments that might be affected by market places. We drawa convex polygon covering all the shopping centers and findthe road segments withing the polygon. In order to draw thepolygons, we drew circles with radius 300 meters with theshopping mall as center and ensured that the polygon coversall those circles. The area size is 4.02 square kilometers. Thisarea contains 11 shopping complexes. The extra space (300

meters radius for each shopping mall) is taken to investigatethe traffic pattern beside the markets. The red points are theshopping centers in figure 4. There are 362 road segments inthat area.

Fig. 4: Markets (red points) and road segments (blue lines) inmarketplaces study area (4.02 square kilometers).

We compare the traffic intensity pattern of the area shown infigure 4 considering average traffic intensity pattern for bothmarket’s closed and the days when markets are kept open.Figure 5 shows the comparison of traffic intensity patternfor market closed and those days when the markets are keptopened.

Fig. 5: : Effect of market places on traffic congestion. The reddashed curve represents the traffic pattern when the marketsare kept open and the blue solid curve is for market closeddays traffic pattern.

We find that traffic pattern does not change much from00:00:00 to 16:00:00. However, high traffic intensity for a longperiod indicates that marketplaces increase traffic congestion.

D. Traffic Intensity Pattern for Rickshaw Free RoadsRickshaws are slow moving pedicabs and has been con-

sidered a primary reason of traffic congestion in Dhaka city.There is policy to make some roads as rickshaw free. Weanalyze the traffic pattern for rickshaw free roads. We collectthe information of rickshaw free roads of Dhaka city fromJICA report [7]. From our data set, we find 22 road segmentswhich are rickshaw free. We take the average intensity of15 days to find the traffic pattern. Figure 6 shows the trafficintensity pattern of rickshaw free roads.

The red curve shows average traffic intensity informationand green line is the standard deviation bar. From figure 6, we

Fig. 6: Rickshaw free roads traffic intensity pattern. Trafficintensity is low after 7:00 pm.

see that the roads are congested most of the time. The averagetraffic intensity for rickshaw free roads decreases from 7 pmas the vehicle pressure is low at that time (not office hour).In general, rickshaw free roads do contribute lessening trafficintensity.

E. Road Segments Intersections and Traffic Congestion

Number of intersections at roads plays an important role intraffic congestion. We analyze the intersections of every roadsegment for each DPZ zone. We use the latitude, longitudeinformation of road segments to find the intersections fromthe data set. If two road segments connect with each otherdirectly then there is an intersection for both of those roadsegments. We use line segments intersection algorithm for thispurpose. As length of road segments differs, we compute thenumber of intersections per kilometer to analyze the effect.Each road segment is categorized into having intersectiondensity (intersections/ kilometer) 0 to 3, 3 to 6, and 6+. Figure7 shows the average traffic intensity for each case.

Fig. 7: : Average traffic intensity for different road segmentsintersections density (intersections/ kilometer). We observe,traffic intensity is high for most the time period where inter-section density 6+ and it is low where the intersection densityis [0-3].

We find that, for road segment with [0-3] intersec-tions/kilometer, the traffic intensity remains low most of thetime. From 1 am to 4 pm the traffic intensity is high for theroad segments with 6+ intersection density.

F. Finding Zone Clusters and Traffic Intensity Pattern

Next we analyze the traffic pattern of Dhaka city in a moremacro level. As mentioned earlier, from RAJUK report [18],we obtain 13 DPZ zones of Dhaka city. We determine the sim-ilarities between different DPZ zones. In order to do this, weuse the land use pattern from RAJUK report [18] to measurethe similarity between zones. We use hierarchical clustering

method to find the clusters. Agglomerative clustering methodhas been used using Euclidean distance between two objectsas distance metric. We take the single linkage criteria andproduce a dendrogram shown in figure 8.

Fig. 8: Dissimilarities among zones using hierarchical clus-tering method using land use information. DPZ 6 is taken asourlier as it is dissimilar from other zones.

Figure 8 shows us that DPZ 6 is the most dissimilar zoneconsidering land use information and DPZ 5 and DPZ 12are the most similar zones. We take four clusters from thatanalysis considering each cluster has the most similar set ofzones. Cluster 1 (a mix of residential and commercial areawith narrow roads) contains DPZ-1 and 2. Cluster 2 (mostlycommercial area) contains DPZ-3 and 7. Cluster 3 (mostlyresidential area) contains DPZ 5, 12 and 13. Cluster 4 (mainlyin suburb area) contains DPZ-8, 9, 10 and 11.

Fig. 9: : Traffic Intensity Pattern for all 4 clusters. Only cluster-2 shows similar traffic pattern among zones.

Figure 9 shows different clusters and the traffic intensityrecords with respect to time. For, cluster 1,3 and 4, trafficintensity varies in zones even if those fall in same cluster.Therefore, land use pattern does not itself explain the variationin traffic intensity.

G. Analyzing Traffic Intensity Factors

Since, only land use pattern dose not explain the variationof traffic pattern, we use other factors with land use patternand perform regression analysis. We find the number of roadintersections (intersections per kilometer) and number of busroutes of all the DPZ zones. Table II shows the result ofour regression analysis. The R square value of this regressionanalysis is 0.880 and the adjusted R value is 0.820. Here, thedependent variable is the average traffic intensity of each zone.

TABLE II: Analyzing factors of land use that influence trafficcongestion

Variables Coefficients Std Error P-valueIntercept 0.378 0.041 1.54E-05% of Road Network Area 0.022 0.004 0.001% of Mixed Land Area 0.002 0.001 0.021# of Intersections/kilometer 0.019 0.005 0.007# of Bus routes -0.002 0.001 0.035

From table II we observe that road network, mixed land areaand road intersections coefficient values are all positive. Thep values are also significant(95% confidence level) in all thecases. However, number of bus routes coefficient is negative.But, the value is small to strongly conclude that increasingpublic transportation will decrease traffic congestion.

In summary, we find that Dhaka city suffers from trafficcongestion in most of the time of a day. Weekdays trafficintensity is higher than weekends traffic intensity. Marketplaces also creates extra traffic intensity. Rickshaw free roadsare free from congestion after 7:00 pm. Though zones insame cluster have similar in land use pattern, traffic intensityvaries among those zones. Road intersections create extratraffic congestion as the traffic intensity increases at the roadsegments with higher intersections. Road network and mixedland area influence traffic congestion (positive coefficient inregression analysis). However, number of bus routes coeffi-cient is negative in regression analysis. Therefore, the priorityshould be assigned in public buses as public transportationmode rather than private rickshaws which create extra trafficcongestion.

V. CONCLUSION

Accurate traffic intensity pattern is useful in order to findthe reasons behind traffic congestion. In our research, wehave analyzed the traffic intensity pattern macroscopically. Wedescribe the traffic pattern for different cases (whole day, effectof marketplaces, day wise traffic variations, traffic pattern ofrickshaw free roads). As the road intersection is a reason oftraffic congestion, we need other alternatives (e.g. fly overs, U-loops) to reduce the intersection points. Public transportationhelps to reduce the traffic intensity as our analysis provides

the negative relation between number of public bus routes andtraffic intensity.

Further research should be done for better traffic modelingin order to find the accurate traffic intensity pattern. Trafficsensors (traffic image, noise detection, GPS devices) shouldbe installed to analyzed the data more efficiently. For mod-eling MFD, vehicle density is required for particular roadintersections. Social infrastructures (residential, educational,commercial, industrial) effect should also be analyzed toreduce traffic congestion. Since, we have the GPS data foronly 15 days, we need more data for better traffic modelingthat can contribute predicting traffic congestion informationand provide route suggestion accurately.

REFERENCES

[1] ADB (2011), “Preparing the greater Dhaka sustainable urban transportcorridor project”, consultants report.

[2] Mansura Hossain, “Dhaka’s traffic congestion costs tk550b yearly”,Prothom-Alo(Newspaper), July 12, 2015. Available online athttp://en.prothom-alo.com/bangladesh/news/71961/Dhaka-s-traffic-congestion-costs-Tk-550bn-yearly. Accessed on : 24th January 2017.

[3] Hossain, M. “Shaping up of urban transport system of a developingmetropolis in absence of proper management setup: the case of Dhaka.”Journal of Civil Engineering (IEB) 32.1 (2004): 47-58.

[4] Malaya Tashbeen Barnamala, “Traffic jam is freezing strong economyand healthy environment: A case study of Dhaka city.” volume 6, pages36-40, 2015.

[5] Dr. Bjorn Lomborg, “The smartest ways to deal with traffic congestion indhaka.” The Daily Star(Newspaper), May 09, 2016. Available online athttp://www.thedailystar.net/op-ed/ politics/the-smartest-ways-deal-traffic-congestion-dhaka-1220956. Accessed on : 1st January 2017.

[6] Md. Saidur Rahman, “The only solution.” The DailyStar(Newspaper), March-2010. Available online athttp://archive.thedailystar.net/forum/2010/march/only.htm. Accessedon : 6th January 2017.

[7] Japan International Cooperation Agency, “Preparatory Survey Reporton Dhaka Urban Transport Network Development Study (DHUTS) inbangladesh. 2010.”

[8] Tsubota, T., A. Bhaskar, and C. Edward. “Brisbane Macroscopic Fun-damental Diagram: Emperical Findings on Network Partitioning andIncident Detection.” Trans Res Rec (2014).

[9] Wang, Yanli, et al. “Reasons and countermeasures of traffic congestionunder urban land redevelopment.” Procedia-Social and Behavioral Sci-ences 96 (2013): 2164-2172.

[10] Alam, M., and Faisal Ahmed. “Urban transport systems and congestion:a case study of indian cities.” Transport and Communications Bulletin forAsia and the Pacific 82 (2013): 33-43.

[11] Geroliminis, Nikolas, and Carlos F. Daganzo. “Macroscopic modelingof traffic in cities.” TRB 86th annual meeting. No. 07-0413. 2007.

[12] Liu, Yu, et al. “Urban land uses and traffic source-sink areas: Evidencefrom GPS-enabled taxi data in Shanghai.” Landscape and Urban Planning106.1 (2012): 73-87.

[13] Vacca, Alessandro, and Italo Meloni. “Understanding route switchbehavior: An analysis using gps based data.” Transportation ResearchProcedia 5 (2015): 56-65.

[14] Castro, Pablo, Daqing Zhang, and Shijian Li. “Urban traffic modellingand prediction using large scale taxi GPS traces.” Pervasive Computing(2012): 57-72.

[15] Yoon, Jungkeun, Brian Noble, and Mingyan Liu. ”Surface street trafficestimation.” Proceedings of the 5th international conference on Mobilesystems, applications and services. ACM, 2007.

[16] RAJUK (Rajdhani Unnayan Kartripakkha). “Dhaka structure plan 2016-2035.”

[17] Hanson, Susan, and Genevieve Giuliano, eds. The geography of urbantransportation. Guilford Press, 2004.

[18] RAJUK (Rajdhani Unnayan Kartripakkha) , “Preparation of detailed areaplan (dap) for dmdp area: Group-c, rajuk.”

[19] Number of registered motor vehicles in dhaka (yearwise), Available on-line at http://www.brta.gov.bd/newsite/en/dhaka-metro-up-to-june-2016/.Accessed on : 2nd January 2017.