Evaluating Location Predictors with Extensive Wi-Fi Mobility Data

Libo Song, Dartmouth College
David Kotz, Dartmouth College
Ravi Jain, DoCoMo USA Labs
Xiaoning He, DoCoMo USA Labs

Abstract

A fundamental problem in mobile computing and wireless networks is the ability to track and predict the location of mobile devices. An accurate location predictor can significantly improve the performance or reliability of wireless network protocols, the wireless network infrastructure itself, and many applications in pervasive computing. These improvements lead to a better user experience, to a more cost-effective infrastructure, or both. Location prediction has been proposed in many areas of wireless cellular networks as a means of enhancing performance, including better mobility management, improved assignment of cells to location areas, more efficient paging, and call admission control. To the best of our knowledge, no other researchers have evaluated location predictors with extensive mobility data from real users. In this poster we compare the most significant domain-independent predictors using a large set of user mobility data collected at Dartmouth College. In this data set, we recorded for two years the sequence of wireless cells (Wi-Fi access points) frequented by more than 6000 users. We found that the simple Markov predictors performed as well or better than the more complicated LZ predictors, with smaller data structures.