Discovering the Precursors of Traffic Breakdowns Using Spatiotemporal Graph Attribution Networks

A traffic breakdown contains phases of trigger&formation (A), propagation (B) and dissipation (C). Our goal is to discover the potential traffic breakdown precursors from region X, which is the downstream area antecedent to the breakdown trigger. Traffic breakdowns, characterized by sudden congestion and reduced vehicle speeds, can lead to severe accidents and increased travel times. Identifying the contributing factors enables the development of predictive models to mitigate these events. Several methods have been developed to identify and predict traffic breakdowns. Statistical estimators and probabilistic models analyze transitional events, with one approach using statistical estimators to assess breakdown probability by classifying these occurrences [1]. Machine learning techniques, such as artificial neural networks, have also shown promise for modeling abrupt traffic transitions [2]. However, a key limitation of current methods is their inability to systematically link environmental and driver behavior factors with the spatiotemporal dynamics of traffic breakdowns. For instance, while studies highlight precursors such as road geometry or the braking of a lead vehicle in a platoon [3, 4], input data is often simplified into tabular formats.