Abstract

Smart city infrastructure relies on extensive interconnected networks to provide smart city services including, among other things, transportation, public safety and automated operation of urban systems. While having these types of connections may provide tremendous efficiency to the smart city manager, it also creates increased vulnerability to complex and ever changing cyber threats. Current methods of intrusion detection primarily analyze network traffic from one location to another in a singular event which hinders their capability of recognizing how the attack spreads through the smart city at different times and from different locations. This work presents a spatio temporal graph neural network (ST-GNN) based framework for smart city network threat. The model produces zone level anomaly scores over time enabling analysis of when and where abnormal network behavior is likely to occur. A prototype demonstration system was built using FastAPI to provide proof of concept for real time threat detection and visualization. Preliminary evaluation suggests the framework captures spatio temporal trends in network traffic and shows promise for providing actionable insights in smart city monitoring contexts, achieving 94.2% accuracy and outperforming baseline models in recall and F1-score.