Differentially private and utility-aware publication of trajectory data

Abstract

Trajectory data is valuable for various applications, especially for intelligent transportation systems, which hunger for plenty of trajectories. However, publishing trajectory data while respecting users’ privacy has been a long-standing challenge. Currently, the prevailing releasing solutions usually merge trajectory locations based on k-means and add unbounded noise with Laplace distribution to the count of trajectory to achieve differential privacy protection. The trajectory merging methods based on k-means have a low efficiency and unbounded noise with Laplace distribution will leak user’ privacy and suffer from serious utility loss. To solve the above two problems, we devise two differentially private and utility-aware publication methods of trajectory data. More specifically, we first propose two trajectory merging schemes based on k-means || clustering. The first one is to use k-means || clustering algorithm to cluster the location area, and all the points in the cluster are replaced by the center of cluster. The other is to utilize Staircase mechanism to perturb the cluster centers in order to improve the level of privacy protection. Afterwards, we propose a bounded Staircase noise generation algorithm to perturb the true count of generalized trajectories. We prove our proposed methods preserve differential privacy theoretically. Experimental comparison show that our proposed publication methods significantly outperform existing approaches in terms of data utility and efficiency.