Activity-Based Human Mobility Patterns Inferred from Mobile Phone Data: A Case Study of Singapore

Transcription

1 Activity-Based Human Mobility Patterns Inferred from Mobile Phone Data: A Case Study of Singapore By: Shan Jiang, Joseph Ferreira, Jr., and Marta C. Gonzalez Published in: 2017 Presented by: Masijia Qiu

2 Author Shan Jiang A Postdoctoral Associate in the Department of Urban Studies and Planning (DUSP) at MIT Joseph Ferreira Founding director of the Planning Department's Computer Resource Lab and is now head of Urban Information Systems Marta C. Gonzalez Works in the area of urban computing Human dynamics, complex systems, network science, data science for the built and the natural environment

3 Current challenges Cities have become more diverse and complex More than half of the global population lives in urban areas (54% in 2014) It is projected that additional 2.5 billion urban population will be added by 2050 The conventional methods: Collects survey data in small sample sizes and low frequencies Requires innovation to link the massive data with the rise of the ubiquitous sensing technologies, digital human footprints

4 Aim of the paper Demonstrates an application of big data analytics Translate ubiquitous mobile phone data into planner interpretable human mobility patterns Activity-based approach With Singapore (a city-state) as an example Understands how individuals travel and conduct activities Extract statistically reliable information Quantify spatial distributions of travel patterns Target specific urban areas to address urban challenges

5 Importance Improve transportation-planning Traffic congestion (peak-hour congestion mitigation) Environmental pollution and degradation Energy consumption Greenhouse emission Sustainable urban development Improve urban economic structure Improve urban mobility and quality of life

6 Study Data (1) Singapore A city-state With a range of 43 kilometer in the west-east direction and 23 kilometer in the north-south direction A population of 5.18 million in million residents and 1.39 million non-residents One of the world s highest mobile penetration rate 150 percent (using total population as base)

7 Study Data (2) Call Detail Record (CDR) data Two consecutive weeks (i.e., 14 days, in March and April of 2011) from one carrier in Singapore 3.17 million anonymous mobile phone users, and a total of million records of phone usages More than 5 thousand cellular towers in Singapore With a spacing gap of around 50 meters to a few kilometers The data from one mobile carrier only count for 63 percent of the total population

8 Study Data (3) Divide census data by demographic group Geographic zones (Development Guide Planning zone, DGP) Individual older than 10 year-old Expand users

9 Study Data (4) Survey Data Singapore 2008 Household Interview Travel Survey (HITS) data collected by the Singapore LTA 34,000 individuals and their 1 day travel information Small sample-sizes Expanding survey samples to total population Compare with estimates from the CDR data To validate results estimated from the CDR data

10 Method: workflow 1. Extract stay location 2. Detect home 3. Filter data 4. Daily mobility network 5. Expand data

11 Method: Extract Stay Point Trajectory data mining Modified the algorithm for the tower-based CDR trajectories Eliminate noises in the raw CDR data Cluster points that are spatially close (within the threshold of distance) and temporally adjacent (in the time order) into a single location

12 Method: Detect Home It is important to label the mobile phone users home location (tower) combine with census data (record population by residential location) To form mobility networks To understand individuals mobility patterns from their home Home tower identification: the most frequently communicated tower during weekdays night (7pm to 7am), and weekends over the study 2.88 million users whose home towers can be observed in the 2-week data 91 percent of all phone users contained in the data for the study period 56 percent of the total population.

13 Method: Filtering Phone User-Day Samples Sample filtering rules: Only keep a user-day observation, if in a day (24 hours) the user has phone records in at least eight distinct time-slots of the 48 half-hour time-slots Separate observations on weekdays from those on weekends 6.28 million user-days for 1.55 million users in the 10 weekday days Filtered users consists of 49 percent of all users in the raw data

14 Method: Identifying Individual Mobility Networks Mobility motifs Directed graph Directed edges

15 Method: Mobile Phone Usage Patterns and Mobility Patterns It indicates that phone users with different phone usage patterns do not have systematic differences in travel behavior. This verification is critical Validity of using active mobile phone user-day samples Of using these samples to expand to the population

16 Method: Expanding Mobile Phone Sample to Population User expansion factors The user expansion factor for every user with home tower in the same planning area is the same Q β i = P g / g 1 U hg User-day expansion factors Normalize the expansion factor by the total number of observation-days for this user. θ ik = β i / K i

17 Method: Aggregate Daily Trips and Mobility Patterns Aggregate the daily trips and mobility patterns for Singapore residents (above 10 years old) at the metropolitan level Motif with two-nodes reported in the survey is dominant (55%) >> CDR data

18

19 Spatial Patterns: The Two-Node Home- Based Tour residents had one additional place besides home Location Quotient Compare an area s distribution of residents by mobility to the distribution of a reference area LQ m = (e m /e)/(e M /E) West region: East, Jurong West, Clementi, Bukit Batok, Choa Chu Kang; North-West region: Woodlands, Sembawang, and Yishun; North-East region: An Mo Kio, Sengkang, and Punggol New towns and far away from the city center Potentially important for public transportation services

20 Spatial Patterns: The One-Node, Three-Node and Four-Node Motifs Weekday stay-at-home residents In the Central region, it contains more mature neighborhoods where senior residents have higher probability to reside. Three- and four-node motifs These areas include suburbs that are connected by light rail

21 Conclusion Traditional travel survey data, although rich in detail, can be misleading The authors present an integrated pipeline that can parse, filter, and expand the raw passive and massive mobile phone CDR data to extract human mobility patterns in a metropolitan area, and translate the knowledge gained into planning-interpretable languages Big Data, if properly treated, can provide further insights beyond travel surveys, improve the transportation-planning and urban-planning

22 Discussion: Strengths CDR data is the longer observation period and larger sample size Applied sample expansion method to expand modeled results from mobile phone users to population at the metropolitan level Weaknesses CDR not always reveal individual travel as travel dairies or GPS tracking applications do Did not talk about the over-hold of the mobile devices (150% mobile penetration rate)

23 Discussion: Related Papers City in big data Mone, G. (2015). The new smart cities. Communications of the ACM, 58(7), Kitner, K. R., & de Wet, T. (2015). Big city, big data. interactions, 22(4), Traffic management and big data Greengard, S. (2015). Smart transportation networks drive gains. Communications of the ACM, 58(1),

24 Discussion: Future Work Further extend the method to examine more details at clusters One- or two-node versus three- or four-node motifs Include understanding the demographics Housing types of the clusters Planners can design policies that target neighborhoods with great potential for improvement Arrange shopping clusters along certain stops for transit-oriented development Add transportation alternatives Improve level of service along certain corridors Integrating land-use transportation planning

25 Discussion Can this method be applied onto data type other than mobile phone? If not, what s the difference? How can we make it possible? If applied on other area/state/country, how will the mobile penetration rate impact the result? Is it possible to include the mobile penetration rate into the analysis? Can this analysis method be extended beyond metropolitan area? What barrier will it meet? How to overcome them?