Detecting Non-Recurrent Congestion in Urban Networks: A Comparative Analysis of Advanced Computational Models and Traffic Pattern Visualization

Traffic congestion, especially non-recurrent congestion (NRC), severely impacts urban mobility and safety. To address this, it is crucial to swiftly identify NRC events, leading to optimized traffic management and fewer disruptions. Utilizing the Hong Kong government’s strategic route data, such as speed, flow, and occupancy parameters, this study seeks to pinpoint and assess NRC instances. We employed a range of computational methods: One-Class Support Vector Machines (One-SVM), recurrent neural networks (RNN), heuristic algorithms, and the Isolation Forest. To validate these models, predictions were compared with official accident records, providing insight into their accuracy in detecting non-recurrent congestion. Additionally, visual representations of congestion data further enriched our assessment. Our analysis reveals the Isolation Forest model’s superior performance in identifying NRC, validated through experiments with accident data. This emphasizes the crucial role of advanced analytics in enhancing urban transport intelligence and safety.