Analysis of location prediction performance of LZ algorithms using GSM Cell-based location data

Abstract

Predictions about users' next locations allow bringing forward their future context, thus having additional time to react. To make such predictions, algorithms capable of learning mobility patterns and estimating the next location are needed. This work is focused on making the predictions on mobile terminals, thus resource consumption being an important constraint. Among the predictors with low resource consumption, the family of LZ algorithms has been chosen to study their performance, analyzing the results drawn from processing location records of 95 users. The main contribution is to divide the algorithms into two phases, thus being possible to use the best combination to obtain better prediction accuracy or lower resource consumption.