CENS Technical Seminar Series

Inferring Mobility States using GSM and Wi-Fi Traces from Mobile Phones

Invited Speaker: Min Y. Mun
Date: October 31, 2008
Time: 1:00 PM - 2:00 PM
Venue: Boelter Hall 4760

Abstract

Inferring mobility states such as being stationary, walking, or driving is critical for transportation studies, urban planning, health monitoring and epidemiology. Our focus is on building a parsimonious mobility classification system using mobile phones with the goal of large deployment, which poses new design considerations - device heterogeneity, processing complexity, energy efficiency, user-time coverage, application diversity. Previous work focuses on fine-grained location-based mobility inference using global positioning system(GPS) data, however, GPS-based mobility characterization raises many issues such as spotty coverage and battery drain that are inadequate to meet our goal.
In this talk, we propose a new mobility classification method using GSM and Wi-Fi traces that are already available on most commercial mobile phones. This method provides users opportunities to use application services while using energy-inexpensive existing infrastructures; Sampling GSM and Wi-Fi data consumes relatively low energies (GSM,Wi-Fi: 0.23 watts, GPS, Accelerometer: 0.424 watts) and Nokia N95 lasts 24.3 hours while capturing theses data. We demonstrate how coarser-grained mobility states can be satisfactorily inferred from this parsimonious approach for some applications and usage models using a data set of 6.25 hours gathered in five differently characterized environments. Furthermore, we show that this model trained from the small data set by one user works well on a data set of twenty hours from sixteen individuals with 78% accuracy, which shows potential in scaling up to other users. Furthermore, we evaluate mobility classification methods using other sensors such as a GPS receiver and an accelerometers sensor to support applications requiring high accuracy (over 90%) or the identification of fine-grained activities.

Biography

Min Y. Mun was born in Seoul, Korea. She received her B.S. in Computer Science from Ewha Women’s University, Seoul, Korea in 2004, and her M.S. in Computer Science from University of Southern California in 2006 under the supervision of Prof. Cyrus Shahabi. She joined The Center of Embedded Networked Sensing last Fall and is currently working on PEIR (Personal Environmental Impact Report) project under Prof. Deborah Estrin and Prof. Mark Hansen. Her research interests include ubiquitous computing and mobile computing using Machine Learning techniques.