1. Enhancing Policy Decision Making with Large-Scale Digital Traces Vanessa Frias-Martinez University of Maryland NFAIS, February 2014

4. 5.9 billion 87% 3.2 billion unique users 45% mobile devices >>humans

5. Have you ever heard of DATIFICATION? 1. Yes 2. No

6. Mobile Digital Footprints… …for Social Good?

7. Research Goal To extract human behavioral information from mobile digital traces in order to assist decision makers in organizations working for social development

8. TOOLS BEHAVIORAL INSIGHTS Energy RESEARCH DECISION MAKERS Health Education Safety Transportation Interviews, surveys: Information to assist on policy decisions Data Mining Machine Learning Statistical MOBILE DIGITAL TRACES To enhance or complement information in an affordable manner

9. OUTLINE

10. Outline Cell Phone Data Projects with Social Impact – Cencell – AlertImpact

11. Cell Phone Data

12. Call Detail Records Anonymized Granularity 1-4km² CDR: Caller | Callee | Date | Duration | Geolocation

13. Modeling Human Behavior Over 270 variables

14. Cost-Effective Census Maps From Cell Phone Data CenCell

15. Motivation: Census Maps A/B C+ C D E

16. National Statistical Institutes A/B C+ C D E

17. Important Data Comes at a Price Expensive Low resource regions A/B C+ C D E

18. Can the variables extracted from Call Detail Records be used as predictors of regional socioeconomic levels (SELs)?

19. Cost-effective Maps NSI carries out surveys Cell Phone Data REDUCE COSTS NSI surveys subset of regions Forecasting Models Predict the Present

20. Methodology

21. Classifying SELs - Training Consumption Social Mobility SEL CLASSIFIER Aggregated 1-4km²

22. SEL Classifying SELs - Testing CLASSIFIER Consumption Social Mobility Aggregated

23. Experimental Evaluation

24. Datasets Data for a city in Latin America (NSI) – 1200 regions (GUs) – SEL values from 0..100 Call Detail Records – 6 months, 500K customers – City has 920 coverage areas – 279 variables per coverage area

25. Evaluation Results Random Forests 86% 3 SELs (A,B,C) EM Clustering 68% 6 SELs (A,B,…,F)

26. Human Behavior and Census Variables

27. Large Scale Quantitative Analysis Consumption Social Mobility

28. Insights Consumption Variables Mobility Variables

29. AlertImpact Understanding the Impact of Health Alerts using Cell Phone Data

30. H1N1 Mexico Timeline Preflu Medical Alert 17th April Closing Schools 27th April Suspension 1st May Reope n 6th May

31. Can we measure the impact that government alerts had on the mobility of the population ?

32. Evaluation Call Records from 1 st Jan till 31 st May 2009 – Compute mobility as different number of BTSs visited Stages – Medical Alert - Stage 1 (17 th -27 th April) – Closing Schools - Stage 2 (28 th -1 st May) – Suspension of Essential Activities - Stage 3 (1 st May-6 th May) Baselines – same periods, different year (2008)

33. Changes in Mobility April 27thMay 1st May 6th AlertClosed Shutdown Reopen Baseline Mobility reduced between 10% and 30% Alert Closed SuspensionReopen

34. Changes in Epidemic Spreading Baseline (“preflu” behavior all weeks) Intervention (alert,closed,shutdown) Epidemic peak postponed 40 hours Reduced number of infected in peak agents by 10% BASELINE K

35. University Campus Statistically Significant Decrease during Stages 2 and 3

36. Airport Statistically Significant Increase during Stages 2 and 3

37. Take Away Message

38. Geolocated traces allow us to quantitatively – Model human behavior – Measure behavioral changes – Predict/Classify external sources of information

39. Future Enhance and complement the tools currently used by decision makers in organizations working for social good – Use of open datasets, social media and other digital traces

40. Thanks !! vfrias@umd.edu