Privacy budgeting for spatio-temporal trajectories
Author Identifier (ORCID)
Abstract
Beneath the surface of location logs lies a wealth of unique behavioral patterns that can betray individual identities, even after conventional anonymization. Despite advances in privacy-preserving techniques, existing solutions fall short by overlooking the quantitative correlations across space and time that define real-world trajectories. Employing the framework of differential privacy (DP), this paper establishes analytically that privacy risks are not isolated, and the exposure at any point in a trajectory is shaped by where it came from and when it occurred. Moreover, this paper reveals the importance of adequate distribution of the privacy budget along a spatio temporal trajectory. Specifically, with the analytical expression established, we formulate the privacy budget allocation as a non-trivial, non-convex, constrained optimization problem, which aims to maximize the utility under the constraint of a privacy budget. We solve the problem judiciously using successive convex approximation (SCA), which not only preserves data utility but also strengthens privacy guarantees by aligning protection levels with the actual sensitivity of different trajectory segments. Experiments demonstrate that our method achieves superior resilience against both traditional and deep learning-based linkage attacks, outperforming existing state-of-the-art solutions.
Keywords
differential privacy, privacy budget allocation, spatio-temporal trajectory
Document Type
Journal Article
Date of Publication
1-1-2026
E-ISSN
15582191
ISSN
10414347
Publication Title
IEEE Transactions on Knowledge and Data Engineering
Publisher
IEEE
School
School of Engineering
Funding Information
National Natural Science Foundation of China (Grant Number: 62401186)
Copyright
subscription content
Recommended Citation
Zhu, S., Sun, C., Yuan, X., Li, C., Ni, W., & Zhang, W. (2026). Privacy budgeting for spatio-temporal trajectories. IEEE Transactions on Knowledge and Data Engineering, 38(10), 6942–6956. https://doi.org/10.1109/TKDE.2026.3718033