CATS: Conditional Adversarial Trajectory Simulation for Privacy-Preserved Data Publication

Keywords: location privacy, deep learning, trajectory data, generative adversarial network, geoAI
Session Type: Virtual Paper Abstract
Day: Tuesday
Session Start / End Time: 3/1/2022 03:40 PM (Eastern Time (US & Canada)) - 3/1/2022 05:00 PM (Eastern Time (US & Canada))
Room: Virtual 21

Abstract

The prevalence of location-aware devices contributes to the explosive growth of large-scale individual-level trajectory data, which provides new opportunities for various studies such as human mobility and public health. However, it also raises public concerns about location privacy issues. In this research, we introduce CATS, conditional adversarial trajectory simulation based on an end-to-end deep learning method. CATS generates synthetic trajectories based on customized conditions as alternatives to original trajectories for data sharing and publication. Such a method can be used for preserving key characteristics from original trajectories while minimize the risk of location privacy leakage. The method is evaluated on two real-world trajectory datasets comparing with existing geomasking-based location privacy protection techniques. The results show that our method can better prevent users from being
re-identified while preserving more characteristics of original trajectories. The model better balances the effectiveness of trajectory privacy protection and the utility for spatio-temporal analysis, which offers new insights into the GeoAI-powered privacy protection.