Lightweight Neuro-Symbolic Anomaly Detection of Traffic

Traffic plays a crucial role in modern life, influencing economy, public safety, environmental health, and overall quality of life. As cities grow and transportation networks become more complex, urban planning and monitoring become an even more difficult task. For this reason, extensive sensor networks are deployed across road infrastructures to collect valuable data on a continuous basis. Real-world traffic data is highly dynamic and complex, which makes accurate understanding, timely and meaningful forecasting, and actionable monitoring of traffic behaviour a significant challenge. This paper proposes a neuro-symbolic workflow for lightweight, real-time traffic anomaly detection, designed to handle diverse traffic conditions effectively. The proposed approach is validated in two distinct case studies: a dense urban traffic corridor in Brussels, Belgium, and a large-scale highway network in the San Francisco Bay Area, USA. While the Brussels dataset offers fine-grained temporal data over an extended period, the San Francisco dataset covers a vast number of monitored locations. The results demonstrate the effectiveness of our method in identifying anomalous traffic behaviour, providing valuable insights for traffic management and decision-making.