Privtraj: a Cloud-Native Privacy-Preserving Framework for Secure Smart City Trajectory Analytics Using Differential Privacy

Abstract

The advent of location-aware gadgets and amenities has resulted in an unprecedented scale of trajectory data being generated, gathered and analyzed in a variety of sectors including urban planning, transportation management, monitoring public health, and corporate analytics. However, trajectory data has inherent spatial-temporal granularity, which poses severe privacy risks. Attackers can leverage linkage attacks, re-identification attacks, and inference attacks to infer sensitive attributes, such as home addresses, medical visits, political affiliations, and daily behavior patterns. In this research, we propose a scalable cloud-native system, PrivTraj, to address these privacy difficulties by a hybrid confidentiality method integrating I-diversity anonymization with differential privacy techniques. The proposed system performs context-aware noise injection where the magnitude of the perturbation is adaptively calibrated based on the linguistic sensitivity of the visited locations. This allows the system to provide stronger privacy guarantees for sensitive locations such as medical centers and government establishments, while preserving data utility for less sensitive venues. The system is built around a distributed processing pipeline using Apache Kafka for real-time data ingestion, Apache Spark for parallel trajectory processing and PostgreSQL for persistent storage, all controlled by an interactive frontend built with React and a backend developed in FastAPI. Experimental evaluation shows that PrivTraj achieves a privacy level higher than 85%, while keeping data usefulness above 70%, with processing latencies less than 15 milliseconds for datasets having 750 trajectory points for 8 simulated users and 10 different geographical categories.