A Visual Analytics Approach to Scheduling Customized Shuttle Buses via Perceiving Passengers' Travel Demands
Qiangqiang Liu, Quan Li, Chunfeng Tang, Huanbin Lin, Xiaojuan Ma, Tianjian Chen
AA Visual Analytics Approach to Scheduling Customized Shuttle Buses viaPerceiving Passengers’ Travel Demands
Qiangqiang Liu * Quan Li ∗ † Chunfeng Tang ∗ Huanbin Lin ∗ Xiaojuan Ma ‡ Tianjian Chen ∗ ∗ AI Group, WeBank, Shenzhen, Guangdong, China † The Hong Kong University of Science and Technology, Hong KongFigure 1:
ShuttleVis includes (A) a dataset loader and data description; (B) overview of car-hailing reimbursement records acrossdifferent departments and descriptions of the departure and arrival time; (C) directional clustering configuration view to help analystsidentify appropriate travel directions; (D) map view to visualize identified directional and regional clustering results, and comparativeranking view that includes (E1) a ranking of shuttle bus stops in terms of (E2) metrics in consecutive regional clusters along onetravel direction, (E3) timetables of selected shuttle routes, and (E4) radar chart showing attribute distributions of selected routes. A BSTRACT
Shuttle buses have been a popular means to move commuters sharingsimilar origins and destinations during periods of high travel demand.However, planning and deploying reasonable, customized servicebus systems becomes challenging when the commute demand israther dynamic. It is difficult, if not impossible to form a reliable,unbiased estimation of user needs in such a case using traditionalmodeling methods. We propose a visual analytics approach to facili-tating assessment of actual, varying travel demands and planning ofnight customized shuttle systems. A preliminary case study verifiesthe efficacy of our approach.
Index Terms:
Human-centered computing—Visualization—Visualization design and evaluation methods * { alfredliu,forrestli,jsontang,bindylin,tobychen } @webank.com † corresponding author ‡ [email protected] NTRODUCTION
As a kind of “Mobility as a Service” (MaaS) [18], shuttle buses thattravel along a fixed route following a pre-determined commutingschedule have been a popular means to move commuters sharingsimilar origins and destinations during periods of high travel demand,due to its advantages of congestion alleviation, environment friendli-ness, and better user experience [19]. However, planning shuttle busroutes in a reliable and cost-efficient manner is nontrivial due to thefollowing three challenges: (1) Unreliable and biased collection oftravel demands.
Shuttle bus service providers mainly rely on onlinesurveys to gather travel demands of potential passengers and furtheraggregate similar demands to generate candidate bus routes. Thisstrategy is inflexible since survey results only represent the views ofrespondents – a limited sample of potential commuters, leading topossible biases in the data that may not well reflect the reality [15].As a result, current shuttle service cannot sufficiently fulfill the needsof many commuters, either because of misaligned schedule or incon-venient drop-off spots. (2) Labor-intensive planning and design.
Itis very tedious, inefficient, and costly to plan customized commutershuttle routes and schedules by manually analyzing the collectedtravel demands from online survey results [18]. It takes a long timefrom the initial data collection to the final deployment of a new a r X i v : . [ c s . H C ] S e p oute. (3) Dynamic and varying travel demands. It is hard to detectchanges in travel demands and patterns and adjust the shuttle serviceaccordingly and timely by only leveraging and manually processingthe results of the online surveys. Furthermore, maintaining a rea-sonable route and schedule requires a comprehensive considerationof a variety of factors that may affect the transit experience. Other-wise, shuttle service providers would have to make judgment callseventually. In this study, we collaborate with domain experts andidentify their primary needs and concerns by first exploiting travelreimbursement data to derive commuters’ needs overlooked by thecurrent shuttle services. Then we enable shuttle service providers toexplore possible travel direction and destination clustering for deter-mining appropriate shuttle bus bounds for different areas and stopswhich optimize the reachability to commuters’ home. Based on theresults, we generate and compare candidate bus routes and schedulesby considering the factors that may affect the transit experience suchas the departure time distribution, drop-off spot distribution, walkingdistances to destinations, and traffic conditions. An in-depth casestudy is conducted to evaluate the efficacy of our approach.
ELATED W ORK
Understanding
Human Mobility Pattern has significant impactsand enables various applications such as urban planning [35], regionfunction analysis [31, 32], hotspots detection [36], driving route rec-ommendation [11, 33], and bus route planning [4, 5, 21, 24]. Bastnaiet al. [4] grouped taxi trips into clusters and identified a route con-necting multiple clusters to maximize the sum of each connected tripcluster without considering other constraints such as time. Chen etal. [5] clustered “hot” areas with dense pick-up/drop-off to identifya candidate bus stop and then derived several rules to automaticallygenerate candidate bus routes. The above work solve issues thatcontain multiple origins and destinations and make certain assump-tions in problem formulation and evaluation such as fixed walkabledistance and threshold for cluster split, which may affect the re-sults and subsequent decisions. We focus on one origin to multipledestinations and allow shuttle service providers to explore traveldirections and destination clusters that optimize walking reachabilityto destinations. Kamw et al. [14] proposed a computational modelto help users study the jointly constrained accessible regions, streetsegments, and Points of Interest (POIs), while our work focuses onthe reachability between selected shuttle stop and home destinations.
Vehicle Routing Problem is intensively studied in urban plan-ning and transportation field [5]. Bus network design determinesroutes and operation frequencies to achieve certain objectives, e.g.,the shortest route, shortest travel time, subject to certain constraints.However, objectives selection should consider both operators’ andpassengers’ requirements which are often conflicting, leading toa design trade-off rather than an optimal solution [30]. Early busnetwork design is mainly based on surveys to get travel demandsand passenger flows [9, 12]. Some work [5, 19] leverage passengerOD flows to find a bus route with a fixed frequency, maximizing thenumber of passengers along a fixed route subject to the total traveltime constraint. Different from purely automatic methods, we allowanalysts to interactively configure travel directions and regional clus-ters based on walking reachability. We also compare different routesand consider factors that may affect transit experience.
Traffic Data Visualization mainly deals with three types of traf-fic data: event-, location-, and movement-based data [28], and vari-ous applications have been developed [1, 2, 7, 8, 10, 13, 17, 22, 23, 25,28,29,34]. For example, Schreck et al. [27] combined automatic dataanalysis and human supervision. Users can monitor and control theclustering process and attain appropriate results. Users can also inter-actively refine clustering results such as excluding sub-clusters froma cluster or dividing a cluster into several smaller sub-clusters [1,25].Similarly, we incorporate human intelligence in the analysis loop tointeractively initialize directional and regional clusters.
BSERVATIONAL S TUDY AND R EQUIREMENTS
We worked with a team of domain experts from the Department ofAccounting and Human Resources at WeBank (an internet bank), in-cluding a chief director (E.1, male, age: 37), a financial director (E.2,male, age: 31), a human resource personnel (E.3, female, age: 26)and a shuttle bus operation manager (E.4, male, age: 35) to identifytheir primary concerns about making plans of customized shuttlebus routes. Typically, they planned customized commuter buses infour stages. First, passengers who are willing to take the bus neededto submit a travel demand survey. Second, E.1 and E.3 collected thetravel demand data and classified the passengers with similar traveldestinations to obtain a candidate customized bus route accordingto OD traveling directions. Third, E.4 made a field survey for eachalternative shuttle route to check road conditions and adjusted theinitial design of the shuttle route accordingly. Finally, the actual trialoperation of the customized shuttle bus route was carried out and therunning time of the route was estimated for scheduling. At present,the number of the customized commuter bus routes during the day-time is 18. However, with respect to the daytime shuttle buses, thenight shuttle buses for the overtime has not been well implemented.The majority of the employees who work overtime (i.e., after 21:30)prefer to take a taxi on their own and the number of employees whovoluntarily submit overtime travel demand questionnaires is verylimited, leading to a large number of unmet potential travel demands.E.2 commented that if they directly follow the regular shuttle busroutes during the daytime for night operation, the routes may be bi-ased due to different demands and traffic conditions. In other words,the conventional perception and processing of the travel demandswould inevitably affect the bus schedules and routes. To sum up, weneed to meet the following requirements:
R.1 Understand Employ-ees’ Travel Demands for Overtime.
Offering night customized busroutes needs to have a clear overview of employees’ travel demandsfor overtime.
R.2 Generate and Compare Shuttle Bus Routes.
The experts required an interactive visual exploration that considersthe factors that may affect the transit experience within a predictabletime duration combining with automatic recommendation of busroutes that connect the origin and a sequence of bus stops to thedestination.
R.3 Compare Candidate Shuttle Bus Stops.
Howto optimize the reachability to commuters’ home destinations anddeliver the best walking experience is a great concern for placingthe candidate customized shuttle bus stops.
ATA P ROCESSING AND S YSTEM O VERVIEW
The internal shuttle bus system for the daytime and reimbursementfinancial system launched at WeBank provided us with severaldatasets, which record three types of information: (1) Urban RoadNetwork Data.
The local urban road network data comprises a di-rected graph of the city of
Shenzhen , of which the vertices representroad intersections and the edges represent roads. Specifically, thisgraph has 253 ,
890 vertices and 314 ,
234 edges; (2) Car-hailing Re-imbursement Data.
We collected 93 ,
050 employees’ car-hailingreimbursement records for overtime with de-identification, whichranges from April 1 st to August 1 st , 2019. Each car-hailing reim-bursement record includes the employee ID , departure time , arrivaltime , place of origin , place of destination , and payment amount .Particularly, the place of origin is the location of the company andthe places of destination are the employees’ residential locations.We first unified similar residential areas into the same locations sincethey may differ in the building No.(s) or entrances of residentialdistricts. We then obtained their geographical representations (i.e.,latitude-longitude) for all the locations through an online map API,followed by a manual calibration; (3) Daytime Shuttle Routes andTimetables. As mentioned, there are 18 shuttle routes in the day-time picking up passengers with fixed timetables. Furthermore, toobtain the traveling time and route from one bus stop to another start-ing at different departure timestamps, we obtain a dataset (D ) y dynamically crawling traffic conditions and the recommendedroutes between two stops. To avoid data biases, we obtained thedata of every weekday and conducted necessary calibration of travel-ing routes and average operation of traveling duration. Specifically,starting from 21:30 with an interval of 5 minutes, we obtained therecommended bus routes from the workplace to the first bus stopand the routes between the first and second stop with an intervalof 1 minute, and so on. Following the criteria of setting up shuttleroutes, i.e., move forward , destination-closer , no zigzag routes [5],we developed ShuttleVis (Fig. 1) and conducted the following stepsto allow shuttle service providers to generate and adjust shuttle busstops, routes, and schedules.
Step 1 (cid:13)
Determine Travel Directions via Clustering Configu-ration View (R.1).
The first step is to determine the travel directionsthrough directional clustering. We use K-means to obtain the initialnumber of travel directions. We determine the value of K by findingthe peak point in the relationship between the cluster number and the
Silhouette Coefficient [26] among the generated clusters. As shownin Fig. 1(C), the x -axis represents the number of travel directionsand the y -axis indicates the value of the Silhouette Coefficient corre-sponding to the number of travel directions. It can be witnessed thatthe
Silhouette Coefficient attains the highest value when the numberof travel directions is 2. However, determining the number of traveldirections faces more practical issues. The distribution range of theangle towards the company of the home destinations may expandtoo much to set up only one route that connects all the destinations.To handle this issue, we visualize the angle distribution of the corre-sponding home destination towards the company as a box plot foreach travel direction, from which the experts could clearly observethe angle distribution in the current condition. In other words, themore concentrated the box plot, the more concentrated the angles ofthe destinations towards the company. Taken together, we choose9 as the directional clustering number in this case which has a rea-sonable angle distribution and distribution of reimbursement recordsin each directional cluster. Based on the above processing, eachcar-hailing record has a unique directional cluster id. The result ofdirectional clustering is shown as the colored parts in Fig. 2(B).
Step 2 (cid:13)
Initialize Shuttle Stops by Regional Clustering forEach Travel Direction via Map View (R.1).
For each travel direc-tion, we use regional clustering to initialize the candidate shuttlebus stop that can cover nearby drop-off spots within one cluster foreach travel direction. Fig. 2(C) shows how we generate the regionalclusters for each directional cluster. We first iterate all drop-off spots N . For each drop-off spot n i , we construct a walking distance matrixwith the dimension of M × M . We predefine the walking distancebetween the two spots in each pair of the drop-off spot destinationsin this matrix is within 1000 meters after discussing with the experts.In other words, M is less than N since the distance between anytwo spots can be larger than 1000 meters. Therefore, each drop-offspot destination corresponds to a set in which the walking distancebetween any two spots is within 1000 meters. We choose the setwith the maximal number of drop-off spots as the first regional clus-ter. Next, we remove all the corresponding drop-off spots withinthis regional cluster and repeat the above procedures. Thus, eachcar-hailing reimbursement record then has a unique directional andregional cluster id. We propose a Voronoi grid-based map viewto help experts understand the identified directional and regionalclustering results. We first construct Voronoi grids [3] as the basefor our map design on the basis of the drop-off spots extracted fromthe car-hailing reimbursement records (Fig. 2(A)). Two reasons areconsidered for choosing the Voronoi grid [6]. First, the polygonin Voronoi is irregular, similar to the real-world terrain. Second,the neighborhood representation is convenient to maintain based onDelaunay Triangulation. After constructing Voronoi grids, we usedifferent visual cues to represent the boundaries of regional clustersthat belong to the same or different directional clustering. Partic- ularly, for a specific Voronoi edge, if the shuttle bus stop on eachside of this edge belongs to different directional clusters, we depictthis edge as a solid line; otherwise, if the shuttle bus stop on eitherside of this edge belongs to a same directional cluster but differentregional clusters, we depict this edge as a dashed line. In other cases,we just remove the edge (Fig. 2(B)). Figure 2: (A) Voronoi grid-based map design on the basis of drop-off spots. (B) Directional clusters highlighted by different colors andregional clusters separated by dashed lines. (C) The procedure forgenerating regional clusters. We string (indicated by red arrows) andlink consecutive shuttle stops to each destination.
Step 3 (cid:13)
Recommend Shuttle Routes and Walking Paths (R.2).
Up to this point, we obtain a list of candidate shuttle bus stopsalong one travel direction via directional and regional clustering.The next step is to recommend a shuttle route that strings thoseconsecutive shuttle stops according to the distance to the workplace.To intuitively represent the shuttle route for each directional clusterand the path from the initialized shuttle stop to its covered homedestinations, we estimate the shuttle routes and walking paths to eachhome destination on the basis of the previously mentioned dataset (D ) . As shown in Fig. 2(C), the black line shows a completeshuttle route and multiple lines between the shuttle stop (blackpoints) and home destinations (grey points) indicate the walkingpaths with color indicating different distance reachability. Step 4 (cid:13)
Refine Shuttle Bus Stops and Routes via Compara-tive Ranking View (R.3).
Although the map view conveys infor-mation about the candidate shuttle bus stops and routes, the domainexperts need to explore the properties of each candidate shuttle busstop to assess its reachability. After discussing with the experts, weconsider the following factors that may affect the transit experienceto evaluate a candidate shuttle bus stop (Fig. 1(E2)). (1) avg dist: the weighted average distance from one selected shuttle bus stop tothe other destinations in the same regional cluster; (2) avg dura: theweighted average walking duration from one selected shuttle busstop to the other destinations in the same regional cluster. Note that weighted means that we consider the passenger number at each drop-off spot when calculating the avg dist and avg dura . (3) reach200: the ratio of the number of car-hailing orders from one shuttle busstop to the other destinations within 200m to all the car-hailingorders in the same regional cluster. The definition also applies to reach400 , reach600 , and so on; (4) dist cost: we obtain cost i bymultiplying the distance from one selected shuttle bus stop i to theother destinations in the same regional cluster by the number of theorders and dist cost is just the accumulative value of cost i .Inspired by EmbeddingVis [16] and Marey’s train schedule [20], igure 3: Experts selected two departure timestamps 21:30 and 21:55 and compared the two routes in terms of (1) different metrics. (2) Theyadjusted the shuttle bus stop from “
HeZhengJinYuan ” to “
PanShanHuaYuan ” and (3) finalized the route that starts at the departure time of 21:55. as shown in Fig. 1(E1), we present each candidate shuttle bus stopas a combined bar, in which the length of a single bar with colorsindicates the normalized metric value of the corresponding shuttlebus stop. We line up all the regional clusters in one travel directionhorizontally and connect selected shuttle bus stops across all the re-gional clusters with a curve which forms one candidate shuttle route.The ranking in each regional cluster is based on the value of a certainmetric. The left and right vertical curves between two regional clus-ters indicate the distance between two R-Clusters and arrival timestarting at different departure timestamps by using the dataset of (D ) , respectively. Experts can select multiple routes (up to three)for simultaneous comparison by adding them to a candidate list andthey would appear in the radar chart (Fig. 1(E4)). Driving dura , driv-ing dist , walk reach800 , walk avg dura , walk avg dist , and nums indicate the driving duration, distance, the ratio of walking reacha-bility within 800m, the average duration and walking distance, andthe number of involved car-hailing records around the correspondingdeparture timestamp of the selected route, respectively. A timetableof the selected routes is shown in Fig. 1(E3), with x -axis represent-ing the arrival time and y -axis representing the regional clusters anddistance to the workplace. Experts can choose different departuretimestamps and ShuttleVis automatically recommends a new route.
ASE S TUDY
We introduce a case conducted by E.1 and E.4, who used
ShuttleVis to study the night customized shuttle bus operation. After loadingthe dataset and setting the number of directional clusters to 9 aspreviously mentioned, they chose directional cluster 3 for furtherinspection. First, they needed to select an appropriate departuretime for the shuttle which could cover as many passengers as possi-ble. They witnessed that there are two peaks corresponding to thedeparture time of 21:30 and 21:55 (Fig. 3(1)), respectively, so theyadded them to the candidate list for exploration. From the radarchart in Fig. 3(1), they observed that although the route starting at21:30 covers more passengers (indicated by the nums axis) than theother route, the driving distance and duration of the route startingat 21:30 are higher than that of the route starting at 21:55, i.e., thedriving distance and the driving duration of the blue route is 30%and 18% higher than that of the orange route, respectively. E.1commented that their company provides extra transportation com-pensation for their employees who work after 21:30 and thought thisfinding makes sense since there should be more passengers callingfor car-hailing at 21:30, which may lead to traffic congestion. Theexperts then suggested that the departure time for the night shuttlebus should be 21:55. After determining the departure timestamp,the experts moved to verify the recommended shuttle stops androutes . Note that
ShuttleVis recommends the default shuttle stopsand routes based on avg dist . They observed that in R-Cluster 4, thesystem recommends “
HeZhengJinYuan ” as the shuttle bus stop butthe experts identified that another drop-off spot “
PanShanHuaYuan ” is located in the middle part in this regional cluster (Fig. 3(2)). Toobtain more information, E.1 clicked on the rectangle of “
PanShan-HuaYuan ” in Fig. 3(2)(b) and found that the avg dist , avg dura , and coverage within 800m is 820m, 9.65min, and 93%, respectively,compared with that of “ HeZhengJinYuan ” of which the avg dist is768m but with a longer avg dura (12.94min) and smaller coverage(76%). E.4 reported that this may be due to footbridge or subway(i.e., shorter distance but longer duration). Therefore, they decidedto change the shuttle stop to “
PanShanHuaYuan ” for this regionalcluster. To further inspect differences between daytime and night,they compared the routes by
ShuttleVis for overtime with thedaytime route . As shown in Fig. 4, the red line indicates the stopsand route by
ShuttleVis while the blue one indicates the daytimestops and route. Generally, the blue route connects every drop-offspot but cannot well cover drop-off spots at night (Fig. 4b). InFig. 4a, although the daytime blue route covers most night drop-offspots, it potentially increases the driving duration.
Figure 4: The route adjusted by
ShuttleVis and that in the daytime.
ONCLUSION AND F UTURE W ORK
In this study, we introduce a visual analytics approach
ShuttleVis to facilitating assessment of actual, varying travel demands andplanning of customized night shuttle buses. It allows shuttle serviceproviders to explore traveling directional and regional clusteringthat optimize the reachability to commuters’ home destinations.Based on the identified directions and bus stops, candidate shuttleroutes and schedules are provided for comparison by considering thefactors that may affect the transit experience. In the future, we shallcontinue perceiving travel demands by considering other passengerflows and it may be necessary to set up multiple shuttle routes forone direction in different periods of time if there are many requests. A CKNOWLEDGMENTS
We thank the anonymous reviewers for their valuable comments.This research was supported in part by HKUST - WeBank JointLaboratory Project Grant No.: WEB19EG01-d.
EFERENCES [1] G. Andrienko, N. Andrienko, S. Rinzivillo, M. Nanni, D. Pedreschi,and F. Giannotti. Interactive visual clustering of large collections oftrajectories. In , pp. 3–10. IEEE, 2009.[2] A. Anwar, T. Nagel, and C. Ratti. Traffic origins: A simple visualizationtechnique to support traffic incident analysis. In , pp. 316–319. IEEE, 2014.[3] F. Aurenhammer and R. Klein. Voronoi diagrams.
Handbook ofcomputational geometry , 5(10):201–290, 2000.[4] F. Bastani, Y. Huang, X. Xie, and J. W. Powell. A greener trans-portation mode: flexible routes discovery from gps trajectory data. In
Proceedings of the 19th ACM SIGSPATIAL International Conferenceon Advances in Geographic Information Systems , pp. 405–408, 2011.[5] C. Chen, D. Zhang, Z.-H. Zhou, N. Li, T. Atmaca, and S. Li. B-planner: Night bus route planning using large-scale taxi gps traces.In , pp. 225–233. IEEE, 2013.[6] S. Chen, S. Chen, L. Lin, X. Yuan, J. Liang, and X. Zhang. E-map:A visual analytics approach for exploring significant event evolutionsin social media. In
IEEE Conference on Visual Analytics Science &Technology , 2017.[7] W. Chen, F. Guo, and F.-Y. Wang. A survey of traffic data visualization.
IEEE Transactions on Intelligent Transportation Systems , 16(6):2970–2984, 2015.[8] D. Chu, D. A. Sheets, Y. Zhao, Y. Wu, J. Yang, M. Zheng, and G. Chen.Visualizing hidden themes of taxi movement with semantic transfor-mation. In , pp. 137–144.IEEE, 2014.[9] T. A. Chua. The planning of urban bus routes and frequencies: Asurvey.
Transportation , 12(2):147–172, 1984.[10] N. Ferreira, J. Poco, H. T. Vo, J. Freire, and C. T. Silva. Visualexploration of big spatio-temporal urban data: A study of new york citytaxi trips.
IEEE transactions on visualization and computer graphics ,19(12):2149–2158, 2013.[11] Y. Ge, H. Xiong, A. Tuzhilin, K. Xiao, M. Gruteser, and M. Pazzani.An energy-efficient mobile recommender system. In
Proceedings of the16th ACM SIGKDD international conference on Knowledge discoveryand data mining , pp. 899–908, 2010.[12] V. Guihaire and J.-K. Hao. Transit network design and scheduling: Aglobal review.
Transportation Research Part A: Policy and Practice ,42(10):1251–1273, 2008.[13] H. Guo, Z. Wang, B. Yu, H. Zhao, and X. Yuan. Tripvista: Triple per-spective visual trajectory analytics and its application on microscopictraffic data at a road intersection. In , pp. 163–170. IEEE, 2011.[14] F. Kamw, S. Al-Dohuki, Y. Zhao, T. Eynon, D. Sheets, J. Yang, X. Ye,and W. Chen. Urban structure accessibility modeling and visualizationfor joint spatiotemporal constraints.
IEEE Transactions on IntelligentTransportation Systems , pp. 1–13, 2019.[15] S. Lefever, M. Dal, and A. Matthiasdottir. Online data collection inacademic research: advantages and limitations.
British Journal ofEducational Technology , 38(4):574–582, 2007.[16] Q. Li, K. S. Njotoprawiro, H. Haleem, Q. Chen, C. Yi, and X. Ma.Embeddingvis: A visual analytics approach to comparative networkembedding inspection. In , pp. 48–59. IEEE, 2018.[17] S. Liu, J. Pu, Q. Luo, H. Qu, L. M. Ni, and R. Krishnan. Vait: A visualanalytics system for metropolitan transportation.
IEEE Transactionson Intelligent Transportation Systems , 14(4):1586–1596, 2013.[18] T. Liu and A. A. Ceder. Analysis of a new public-transport-service concept: Customized bus in china.
Transport Policy , 39:63–76, 2015.[19] Y. Lyu, C.-Y. Chow, V. C. Lee, J. K. Ng, Y. Li, and J. Zeng. Cb-planner: A bus line planning framework for customized bus systems.
Transportation Research Part C: Emerging Technologies , 101:233–253,2019.[20] E.-J. Marey.
La m´ethode graphique dans les sciences exp´erimentaleset particuli`erement en physiologie et en m´edecine . G. Masson, 1878.[21] S. A. Oluwadare, I. P. Oguntuyi, and J. C. Nwaiwu. Solving school busrouting problem using genetic algorithm-based model.
InternationalJournal of Intelligent Systems and Applications , 10(3), 2018.[22] H. Piringer, M. Buchetics, and R. Benedik. Alvis: Situation awarenessin the surveillance of road tunnels. In , pp. 153–162. IEEE, 2012.[23] J. Pu, S. Liu, Y. Ding, H. Qu, and L. Ni. T-watcher: A new visualanalytic system for effective traffic surveillance. In , vol. 1, pp.127–136. IEEE, 2013.[24] A. S. M. Rahman, P. Magalingam, N. B. Kamaruddin, G. N. Samy, andS. Perumal. Graph analysis study of a city bus transit network.
Journalof Physics Conference Series , 1551:012004, 2020.[25] S. Rinzivillo, D. Pedreschi, M. Nanni, F. Giannotti, N. Andrienko,and G. Andrienko. Visually driven analysis of movement data byprogressive clustering.
Information Visualization , 7(3-4):225–239,2008.[26] P. J. ROUSSEEUW. Silhouettes : a graphical aid to the interpretationand validation of cluster analysis.
Journal of Computational & AppliedMathematics , 20, 1987.[27] T. Schreck, J. Bernard, T. Von Landesberger, and J. Kohlhammer.Visual cluster analysis of trajectory data with interactive kohonen maps.
Information Visualization , 8(1):14–29, 2009.[28] F. Wang, W. Chen, F. Wu, Y. Zhao, H. Hong, T. Gu, L. Wang, R. Liang,and H. Bao. A visual reasoning approach for data-driven transport as-sessment on urban roads. In , pp. 103–112. IEEE, 2014.[29] Z. Wang, M. Lu, X. Yuan, J. Zhang, and H. Van De Wetering. Visualtraffic jam analysis based on trajectory data.
IEEE transactions onvisualization and computer graphics , 19(12):2159–2168, 2013.[30] D. Weng, R. Chen, J. Zhang, J. Bao, Y. Zheng, and Y. Wu. Pareto-optimal transit route planning with multi-objective monte-carlo treesearch.
IEEE Transactions on Intelligent Transportation Systems , 2020.[31] Y. Wu, D. Weng, Z. Deng, J. Bao, M. Xu, Z. Wang, Y. Zheng, Z. Ding,and W. Chen. Towards better detection and analysis of massive spa-tiotemporal co-occurrence patterns.
IEEE Transactions on IntelligentTransportation Systems , 2020.[32] J. Yuan, Y. Zheng, and X. Xie. Discovering regions of different func-tions in a city using human mobility and pois. In
Proceedings of the18th ACM SIGKDD international conference on Knowledge discoveryand data mining , pp. 186–194, 2012.[33] J. Yuan, Y. Zheng, X. Xie, and G. Sun. T-drive: Enhancing driv-ing directions with taxi drivers’ intelligence.
IEEE Transactions onKnowledge and Data Engineering , 25(1):220–232, 2011.[34] W. Zeng, C.-W. Fu, S. M. Arisona, and H. Qu. Visualizing interchangepatterns in massive movement data. In
Computer Graphics Forum ,vol. 32, pp. 271–280. Wiley Online Library, 2013.[35] Y. Zheng, Y. Liu, J. Yuan, and X. Xie. Urban computing with taxicabs.In
Proceedings of the 13th international conference on Ubiquitouscomputing , pp. 89–98, 2011.[36] Y. Zheng, L. Zhang, X. Xie, and W.-Y. Ma. Mining interesting locationsand travel sequences from gps trajectories. In