Privacy-Aware Synthetic Trajectory Generation for Taxi Mobility Analytics and Anomaly Detection

• Dhanush Gopal Battina

  Author

Emerging applications of urban mobility data create potential new applications of transport analytics and anomaly detection, and pose new threats of privacy leakage. The inclusion of sensitive spatial and temporal data in taxi trajectories raises privacy concerns. In this paper, we present a novel framework, Diff Traj --LM--TAD, for privacy-aware taxi trajectory generation and anomaly detection. We incorporate a diffusion-based generative model to generate human-like real-world trajectories, with a transformer-based language model to detect anomalous movement behavior patterns. In addition, we propose an iterative refinement procedure for anomaly diversity and interpretability, achieved by unsupervised anomaly mining and rule-based anomaly categorization. Further, to balance utility and privacy leakage, we adapt differentially private stochastic gradient descent (DP--SGD) to the generative model training process, and develop a trajectory-level anonymization scheme to protect trajectory identity during anomaly detection. Finally, to validate the capability of our framework, we present experimental results of applying our framework to trajectory datasets from the cities of Beijing, Chengdu and Xi'an. Experimental results show an F1-score of approximately 0.78-0.81 for taxi anomaly detection in these three cities and a Jensen-Shannon divergence of less than 0.05 for all three cities, indicating that taxi trajectories from our framework can preserve important spatial properties. Privacy attack experiments indicate over 70% attack success reduction against the original attacks. Our experimental results suggest the generated taxi trajectories have utility for transportation mobility analytics and anomaly detection, with only a small portion of location information being exposed for anonymized user identities. Our proposed framework for privacy-aware anomaly detection offers a promising future direction for privacy-preserving urban mobility analysis.

• PDF

Copyright (c) 2026 International Journal of Intelligent Systems and Data Science

This work is licensed under a Creative Commons Attribution 4.0 International License.