A Hybrid Reinforcement Learning-Based Method for Generating Privacy-Preserving Trajectories in Low-Density Traffic Environments

Intelligent Transportation Systems (ITS) optimize road network capacity, monitor traffic flow, and enhance overall road safety by analyzing real-time trajectory data. However, the utilization of such data raises privacy concerns, enabling potential attackers to gain insights into users’ real-time activities and personal information. Furthermore, existing privacy preservation methods have multiple limitations, particularly in low-traffic density environments. To address these issues, this paper presents a novel approach for generating realistic trajectories that evade tracking. Existing trajectory generation mechanisms are coarse-grained and cannot adequately preserve the quality of location-based services while safeguarding individual privacy. To overcome this limitation, we first use differential privacy to determine a location near the actual destination and employ a path search algorithm to extract relevant road information. Subsequently, by leveraging our hybrid reinforcement learning model, we generate trajectories leading to this fictitious point. The comparison conducted on real-world maps with other trajectory generation methods reveals its superior ability to preserve spatio-temporal features. Finally, we propose two approaches that use the generated trajectories to protect privacy, ensuring both individual privacy protection and the utility of data.