LP-HMM: Location Preference-Based Hidden Markov Model

Abstract

Lots of mobility data have been generated with the emergence of smart devices and location-based services. The prediction of user mobility has become a key factor driving the rapid development of many location applications. Location prediction has attracted more and more attention in various fields, and many location prediction algorithms have been proposed. The data currently used for researches has many problems such as data noise and redundancy. Many researches directly used raw data and did not consider spatiotemporal characteristics of historical data enough, which leads to low prediction accuracy. This paper proposes a point-of-interest discovering algorithm, which fully considers spatiotemporal characteristics of data. By combining the location preference of users for location with the Hidden Markov Model (HMM), we propose LP-HMM (Location Preference-based Hidden Markov Model), a location prediction model based on location preference and HMM. The proposed model is compared with other location prediction models driven by the massive and real mobile dataset Geolife. The experiment results show that the prediction accuracy of the proposed model can achieve 6.4% and 7% higher than Gaussian Mixture Model (GMM) and traditional HMM respectively.