Brownian bridge model for high resolution location predictions

Abstract

Given a person's current and historical traces, a useful yet challenging task is to predict the future locations at high spatial-temporal resolution. In this study, we propose a Brownian Bridge model to predict a person's future location by using the individual's historical traces that exhibit similarity with the current trace. The similarity of the traces with the current trace is evaluated based on the notion of edit distance. The predicted location at the future point in time is a weighted result obtained from a modified Brownian Bridge model that incorporates linear extrapolation. Both Brownian Bridge and linear extrapolation aim to capture aspects of the individual's mobility behaviors. Compared to using either historical records or linear extrapolation method alone, the proposed location prediction method shows lower mean prediction error in predicting locations at different time horizons. © 2014 Springer International Publishing.