DPTraj-PT: A Differentially Private Spatiotemporal Trajectory Synthesizer Using Public Region Traveltime

Spatiotemporal trajectories are essential for many applications, including traffic analysis, and epidemic prevention and control. However, due to the concerns over individual privacy leakage, the release of trajectory datasets has been challenging. This paper presents DPTraj-PT, a differentially private spatiotemporal trajectory synthesizer using public region traveltime. Different from existing methods which add noise to all the temporal data under differential privacy (DP), DPTraj-PT only adds noise to the trajectories' start time, and uses public region travel-time for temporal data generation. This allows DPTraj-PT to save some privacy budget and hence enhance the data utility. Specifically, DPTraj-PT represents each raw spatiotemporal trajectory by a start spatiotemporal cell and adjacent spatial cells, and adopts a model composed of a prefix tree and a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$q$</tex>-order Markov process for trajectory organization. After adding noise to the model, DPTraj-PT samples trajectories from the noisy model and calculates the temporal data for each point using the public region travel-time. The components of the model also help to improve its resistance to noise. DPTraj-PT not only provides privacy protection, but also retains the essential characteristic of real trajectories. Our extensive experiments on real-life datasets show that DPTraj-PT provides much better data utility than state-of-the-art methods. Our code is available at https://github.com/wys2323/DPTraj-PT.