Novel trajectory privacy protection method against prediction attacks

Abstract

When the public uses Location-based Services (LBSs), their location information is constantly exposed. Owing to the spatiotemporal correlation of trajectories, it is easy for attackers to use historical trajectories and background knowledge to predict future locations of target users. We refer to this type of inference attack as a trajectory prediction attack. To address such potential but threatening attacks in a continuous location query, we propose a novel trajectory privacy protection method. The proposed algorithm aims to generate an indistinguishable perturbed location that is robust to the prediction attack, wherein the user’s real location can be replaced by a perturbed location when submitted to an untrusted server. First, a hidden Markov model-based trajectory prediction mechanism is proposed to simulate predictive attacks and compute the predictability of positions before the trajectory is released. Second, the w sliding window mechanism is designed to dynamically adjust the privacy protection degree of each location point in the trajectory according to the predictability of the location and privacy needs of users. Finally, we propose a bounded noise-adding algorithm based on the Laplace mechanism to improve the usability of data. In our experiments, mutual information, trajectory root-mean-square error, query error, and root-mean-square predictability were used as evaluation criteria, and the performance of the proposed method was comprehensively evaluated. The results show that our algorithm can reduce the trajectory predictability to 0.21 without reducing data availability, which is effective against prediction attacks.