TOWARDS TRANSIT TRIP ITINERARY INFERENCE FROM SMARTPHONE DATA: A CASE STUDY FROM MONTREAL, CANADA

Transcription

1 TOWARDS TRANSIT TRIP ITINERARY INFERENCE FROM SMARTPHONE DATA: A CASE STUDY FROM MONTREAL, CANADA Amir Zahabi Postdoc Fellow Zachary Patterson Associate Professor APRIL6, 2016 GIS-T, RALEIGH, NC

2 INTRODUCTION The workhorse for urban travel data collection has long been household travel surveys. Problems: significant costs, decrease response rate, data quality issues. A myriad of emerging technologies are being developed. Fertile area of research: use of mobile phones for data collection. 2

3 INTRODUCTION With respect to data collected using smartphone apps : stop detection, trip breaking, travel mode inference, travel time estimation, congestion detection, and real time transit tracking. To our knowledge, methods aiming to infer transit itinerary used during transit trips do not seem to have received much attention. 3

4 RESEARCH OBJECTIVES Transit itinerary inference using data collected via smartphone travel survey application, DataMobile. 4

5 RESEARCH OBJECTIVES Infer transit route itineraries combining: -Smartphone-collected GPS -GIS data on transit routes The main purpose of the research: -Establish the ambiguity of the above step for different treatments. 5

6 RESEARCH OBJECTIVES Pj GPS point Bus line 1 Bus line 2 Bus line 3 Pi Transit Trip Itinerary Inference AMBIGUITY 6

7 RESEARCH OBJECTIVES How often does this kind of situation exist? How do we figure out the mode (metro, bus and walk) and the bus line used? 7

8 STUDY AREAS AND DATA Montreal (metro area) 4.02 million people Largest metro region in Quebec(4,258 km 2 ) Transit score of 77. 8

9 METHODOLOGY Establishing the degree of ambiguity: (1)GPS and other data collection; (2)transit trip data extraction; (3)calculating transit route ambiguity under different assumptions, rules and a simple transit route inference algorithm; and (4)evaluating overall transit route ambiguity 9

10 METHODOLOGY GPS Data Collection and Other Data Sources GPS data related to transit trips were collected as part of a travel survey using the smartphone application DataMobile people completed the sur vey and had location data -1,650,536 location points. GIS data: : transit and road network data, metro station locations, location of Concordia campuses, and postal code shape file of Montreal. 10

11 METHODOLOGY DataMobile screenshot, Android DataMobile screenshot, ios 11

12 METHODOLOGY Collected data screenshot, Android and ios (left to right) 12

13 METHODOLOGY Montreal transit network 13

14 METHODOLOGY Only trips which the declared mode was transit was kept. Trip identified: simple trip-breaking algorithm. Home-school based trips done to and from the Island of Montreal were selected. To locate the home location, we compared the declared postal code of residence against the first and last points of individual s daily trips. 14

15 METHODOLOGY Home-based Concordia GPS trip points and transit lines 15

16 METHODOLOGY Algorithm outlining the transit route inference procedure 16

17 ANALYSIS Transit route ambiguity is equal to proportions of trip s distance for which the transit line was ambiguous. Trip GPS points and the bus line buffer analysis 17

18 RESULTS 18

19 RESULTS & SUMMARY STATS To better understand the output of the transit ambiguity processing, we present an example trip. -Visual representation of data fed to the algorithm - An output table similar to the table presented. 19

20 RESULTS & SUMMARY STATS Visual representation of a sample trip s GPS points and possible transit options 20

21 RESULTS & SUMMARY STATS Output of final transit ambiguity processing for sample user Point ID Bus lines in vicinity Proposed Bus Line(s) Walk Metro

22 RESULTS & SUMMARY STATS To see the effect of the algorithm on this sample trip, we report the ambiguity at each step. Baseline ambiguity level: 77% ambiguity, After bus route processing: 10% ambiguity Factoring the walk and metro section of the algorithm too, the ambiguity goes down to only 9%. 22

23 RESULTS & SUMMARY STATS Summary statistic of the trip data and relative baseline ambiguity Trip Distance (km) Trip time (min) Baseline Final Ambiguity (%) Ambiguity (%) Average Min Max Total The table also shows summary characteristics of the transit trips themselves with an average 7.4 km, a min of 1.1 km and a maximum of 29 km. 23

24 HIGHLIGHTS Ambiguous and non-ambiguous proportion of trips Processing Stage % ambiguity % 2 lines % 3 lines % > 3 lines Baseline 72% 11% 5% 56% Bus route 30% 15% 10% 5% Bus route + metro 24% 13% 9% 2% Bus route + metro + walk 23% 13% 8% 2% Final ambiguity processing 4.65% 4.06% 2% 1% Baseline processing: before doing any analysis on the data and simply focusing on the level of non-ambiguous points before bus route, metro, or walk processing. Bus root processing: after the bus line selection portion of the algorithm, without walk and metro processing. Final ambiguity processing: percent ambiguity is reported after processing for walk and metro trips and bus route and time. 24

25 CONCLUSION AND FUTURE IMPROVEMENTS The methodology used in this paper for calculating transit route ambiguity follows a set of GIS-based analysis and processing with different software packages. The calculations demonstrate that in Montreal (dense city), transit route ambiguity without any processing is quite high. The analysis can be improved by better trip breaking algorithm as well as the directionality of the roads could help identify walk segments of the trip 25

26 CONCLUSION AND FUTURE IMPROVEMENTS Caution must be used in the strength of the conviction when reporting these results. Final ambiguity processing reduces percent ambiguity from 72% to 4.65%, that is by 67% overall. Our calculations are likely to be upwardly biased. Therefore, these simple processing rules and algorithms that we have used show potential to help in transit itinerary inference. 26

27 ACKNOWLEDGEMENTS I would like to acknowledge the financial suppor t provided by: FQR-NT under the post-doctoral fellowship program scholarship. FQR-SC Nouveaux chercheurs program Canada Research Chairs Program Canadian Foundation for Innovation Concordia University postdoctoral topup program bursaries which made this research possible. STM and Tram (Transportation Research at McGill) for providing us with the GIS data necessar y for this research 27

28 THANK YOU! QUESTIONS? 28