Protecting trajectory privacy: A user-centric analysis

The existing evaluation methods for location privacy protection mechanism (LPPM) focus mainly on location privacy at the time of issuing a location-based query, and rarely consider trajectory privacy from where multiple successive queries are issued. This paper provides a user-centric analysis of trajectory privacy protection against tracking attacks in presence of quality-loss and energy constraints, focusing on impact of user query behavior within a certain period of time. We consider an attacker with adversarial knowledge on users' activity and mobility profiles to run tracking attacks, against which users employ LPPMs that incorporate location semantic diversity to maximize trajectory privacy. The attack and defense problems are formalized as a Bayesian Stackelberg game for quantitative analysis. Real location traces are used to assess the LPPMs' effectiveness, and to find optimal query strategy. The results confirm that query behavior has a significant impact on improving trajectory privacy. Despite increased number of queries to improve trajectory privacy, the effectiveness may be decreased with a longer observation time. At the same query rate, prolonging observation time is beneficial for privacy improvement. We also find that dummy-based LPPM often provides better protection with a stringent energy constraint, while precision-based LPPM performs better under a stringent quality-loss constraint or under a loose energy constraint. At optimal trajectory privacy, quality-loss of precision-based LPPM is lower than that of dummy-based LPPM, but there is an opposite result in terms of energy cost.