A Benchmark Data Governance Framework for Federated Urban Trajectory Mining under Privacy Constraints
Main article
Abstract
Urban trajectory data provide important evidence for traffic management, public service planning, mobility-risk analysis, and platform operations. However, benchmark development for trajectory mining remains difficult because raw movement traces are sensitive, client devices are heterogeneous, and privacy-preserving learning protocols often report accuracy without comparable governance evidence. This study proposes FedTraj-Gov, a benchmark data governance framework for federated urban trajectory mining under privacy constraints. The framework organizes trajectory benchmarks around seven governance domains: provenance, preprocessing, federated partitioning, privacy control, communication efficiency, utility validation, and audit evidence. Using a design-based evaluation calibrated to public urban mobility datasets and simulated non-IID client partitions, the study compares centralized mining, basic federated learning, differentially private federated learning, secure aggregation with compression, and the proposed governance-oriented benchmark protocol. The results indicate that a privacy aware benchmark can preserve competitive predictive utility while improving reproducibility, reporting completeness, and residual-risk transparency. The analysis also shows that the strongest benchmark setting is not the setting with the highest accuracy, but the setting that achieves a defensible balance among privacy, utility, communication cost, and auditability. The contribution of this article is a practical benchmark governance architecture that turns federated trajectory mining from a model performance exercise into a verifiable data management process for smart-city analytics.
