The 16 th TRB National Transportation Planning Applications

The 16 th TRB National Transportation Planning Applications Conference 16 -277 Deriving Activity Locations Inferred from Smartphone Data Eazaz Sadeghvaziri Ph. D. Candidate, Research Assistant Dr. Xia Jin Ph. D. , AICP, Assistant Professor 1 May 2017

AGENDA ü INTRODUCTION ü METHODOLOGY ü DATA PROCESSING ü SURVEY IMPLIMENTAION ü TRAVEL PATTERN ANALYSIS ü CONCLUSIONS 2

INTRODUCTION Background ü Detailed travel information from persons and households are critical to the understanding of individual’s travel behavior to support transportation planning decisions. ü To support the increasingly complex planning activities, many agencies are facing the challenges of obtaining highly nuanced travel behavior data while managing shrinking financial resources. 3

INTRODUCTION Background ü Traditionally, these data were obtained from household travel surveys, which suffered from low respondent rates, high respondent burden and significant costs for survey implementation. ü New technologies such as GPS and CDR 4

INTRODUCTION Passively Collected Data ü Coordinates and timestamps ü Research showed promise of passive data, such as CDR data, GPS devices and smartphones, to replace traditional household surveys. ü Privacy: The actual ability of these data to accurately capture human mobility patterns has caused public concern to grow. By replacing a participant’s user name with a number records became anonymous. 5

INTRODUCTION Global Positioning System Data ü GPS has a vertical accuracy of 5 meters and horizontal accuracy of 3 meters; 95% of the time. ü Highest: When device is in clear view of satellites like in an open field ü Lowest: When obstructed; between high-rise buildings or in tunnel ü GPS device can be fixed to vehicle. ü Participant was asked to carry device daily. Participant may forget to charge the device or leave it at home. ü Cost of purchasing the GPS units and 6 administering the survey

INTRODUCTION Mobile Network Data ü Data are produced whenever a subscriber uses a cellphone: calling, sending a text, or browsing the internet. ü Time of the activity that triggers the recording and users’ location. ü Due to the proliferation of cellular phones, a large sample of data could be obtained at a minimum cost. ü Low number of available towers ü Tower switching ü No information is provided when the phone is inactive 7

INTRODUCTION Mobile Phone GPS Data ü This method combines advantages of high accuracy from GPS and high penetration rate and low response burden from mobile devices. ü Another advantage is its high accuracy in trip mode detection. ü Ansari (2015); An application was installed on participants’ mobile phone. There were 35 participants over 2 weeks; 6: 00 AM - 9: 00 PM. Transportation modes classified with 96% accuracy. 8

INTRODUCTION Trip Ends Identification ü Trip ends were detected when speed was 0 or very low. ü The amount of time spent at a specific location was calculated as the difference between departure and arrival time. 9

INTRODUCTION Research Goal and Objectives ü Goal: Explore the potential of using smartphone GPS data to advance the understanding in mobility and travel behavior ü Objectives: 1) Feasibility of Using Smartphones instead of traditional household travel survey and develop algorithms and procedures to derive travel information from smartphones 2) Identifying applications in mobility and travel behavior studies 10

METHODOLOGY Google Location History Data ü There were many different apps that could facilitate the retrieval of GPS data from smartphones. ü Google Location History (GLH) was selected for this study for economic and ease of access. This data were completely free to record and sufficiently accurate. Recording the data did not require installing additional apps, provided that the participant had a Google account; participants simply had to allow Google to track their locations. 11

METHODOLOGY Google Location History Data ü Google’s location service uses all available means including Wi-Fi Positioning System (WPS), GPS satellite, and mobile networks to locate the device. GPS helps provide the device with a precise location. ü When the phone is still, data were recorded less frequently and ranges from 1 to 5 minutes; when moving, data were usually recorded every 3060 seconds. It was common to record more than 1, 000 data points per day. 12

METHODOLOGY Survey Implementation ü 46 participants were recruited to participate in the study for 2 months. ü Participants were asked to activate the GLH of their smartphone. ü Participants were checked frequently and finally, they sent their data. 13

DATA PROCESSING ALGORITHM Raw Data � ü FIU EC Coordinate: 14 Lg: -80. 368 Lt: 25. 770

DATA PROCESSING ALGORITHM Sample Record ü First Trip (FT) took place from Home (H) to Work (W) at around 10 AM and Second Trip (ST) was from W to H at around 4 PM. In both trips there was a quick stop at Traffic Light (TL). 15

DATA PROCESSING ALGORITHM 16

DATA PROCESSING ALGORITHM Fix Location Detection ü A fixed location was identified if the distance between two consecutive points was less than 150 feet, and the dwell time in the location was more than 6 minutes. ü 150 feet was chosen as the criteria due to the consideration that a person could move around the office or in the house. 17

DATA PROCESSING ALGORITHM Trip Purpose Identification ü Three types of fixed locations; Home, Work, and Others. ü MATLAB; Criteria ü GIS; land use 18 H W

DATA PROCESSING ALGORITHM Data Validation B 2 B 3 ü Google Map ü GLH ü GPS B 1 B 4 B 5 Google Maps 19 GLH GPS Benchmark Latitude Longitude 1 25. 770865 -80. 368721 25. 7708099 -80. 3687204 25. 770850 -80. 368742 2 25. 840476 -80. 369839 25. 8406045 -80. 3702609 25. 840492 -80. 369838 3 25. 839715 -80. 314308 25. 8398448 -80. 3143386 25. 839730 -80. 314322 4 25. 770350 -80. 313627 25. 7703427 -80. 3136940 25. 770350 -80. 313592

SURVEY IMPLIMENTAION ü Survey Design ü Participants Recruitment 20

TRAVEL PATTERN ANALYSIS ü The result of location detection was suggested an improvement on similar location detection studies (Deng and Ji, 2010). 93. 48% of home and 70. 27% of work places were detected correctly. 21

TRAVEL PATTERN ANALYSIS Time-of-Day Dependence for Long-Distance Trips ü A long. O distance trip is defined for this study as 15 mi or longer one way. H o W 22 Arrival Time Departure Time Location Latitude Longitude 00: 00 07: 01 Home 25. 772 -80. 370 07: 55 10: 49 Other 25. 911 -80. 141 11: 37 18: 34 Home 25. 772 -80. 370 18: 51 19: 50 other 25. 772 -80. 370 20: 37 22: 14 Work 25. 770 -80. 367

TRAVEL PATTERN ANALYSIS ü Two-Dimension Diagram of Longitude and Latitude ü Two-Dimension Diagram of Longitude and Time 23

TRAVEL PATTERN ANALYSIS ü Three-Dimension Diagram of Longitude, Latitude, and Time 24

TRAVEL PATTERN ANALYSIS ü Schematic Example of Time Spent in Different Locations 25

TRAVEL PATTERN ANALYSIS ü Long Distance Trip Frequency of Different Time of Day ü The “ANOVA”, and “Bonferroni” tests indicated that F=3. 0701 which is more than 1, and P-val for testing equality of means is 0. 003 which is less than 0. 05, indicating strong evidence that frequencies of long distance trip are different across time of the day, at 0. 05 level of significance. Also, it was found that, the frequency of long distance travel is significantly 26 higher in the Evening than that in the Morning Peak (P-val= 0. 018).

TRAVEL PATTERN ANALYSIS ü Long Distance Trip Frequency of All Days of Week ü Long Distance Trip Frequency of Different Days of Week 27

TRAVEL PATTERN ANALYSIS ü An Irregular trip is defined for this study as 500 mi or longer one way. As this study was done in Miami, 500 mi means a trip to out of Florida state. ü Irregular Trip Frequency of Different Time of Day 28

TRAVEL PATTERN ANALYSIS ü Irregular Trip Frequency of All Days of Week ü Irregular Trip Frequency of Different Days of Week 29

TRAVEL PATTERN ANALYSIS Demographic Analysis 30

CONCLUSIONS ü This study shows that GLH provides sufficient high resolution data that can be used to study people’s movement without respondent burden, and potentially it can be applied to a large scale study easily. ü The developed algorithms in this study work well. However, due to the limit of the pilot data, analysis conducted in the study cannot be used to draw any conclusions. ü This pilot study shows the potential of utilizing mobile phone GPS data as a supplement or complement to conventional data. ü Given the high penetration of smartphones and the low respondent burden, this data can be collected at large scales for longer periods of time with a low cost. ü This data provide the opportunity to facilitate the investigation of various issues, such as less frequent long-distance travel, daily variations in travel 31 behavior, and human mobility pattern in large spatio-temporal scale.

Thank you for your attention! Contact: esade 003@fiu. edu 32