APU-TrajGen: Adaptive Privacy and Utility Preserving Real-Time Synthetic Trajectory Generation

The increasing dependence on Location-Based Services (LBS) raises significant concerns about protecting user privacy while maintaining data utility. Among emerging solutions, synthetic trajectory generation offers a promising approach to enable data sharing without exposing sensitive location information. This paper introduces APU-TrajGen, an adaptive privacy and utility-preserving method for real-time synthetic trajectory generation. Built on an LSTM-based model, the approach dynamically generates synthetic points directly on the user’s device. The method ensures that the generated trajectories meet the targeted privacy-utility balance. Extensive experiments conducted on the Porto Taxi dataset demonstrate the model’s effectiveness in preserving global, trajectory-level, and semantic-level data utility while providing data privacy. Compared to state-of-the-art methods, APU-TrajGen provides a more flexible and balanced trade-of between privacy and utility, enabling the control over privacy levels without compromising analytical value. Furthermore, the decentralized, real-time approach reduces the risks associated with central data collection. Overall, APU-TrajGen advances the state of the art in privacy-preserving trajectory generation by supporting efficient point-by-point data release, reducing computational overhead, and removing the dependency on trusted third-party processors.