Research

This study presents a novel traffic incident impact analysis framework that addresses the limitations of conventional static approaches, which evaluate post-accident congestion solely through instantaneous speed drops or delay levels. Rather than treating incident-induced congestion as a point-in-time phenomenon, the proposed method interprets it as a dynamic spatio-temporal process encompassing transition, propagation, and recovery. Using SUMO-based microscopic simulation data (speed, density, occupancy, and flow at 1-second intervals), the framework operates in two integrated stages. At the macroscopic level, an Isolation Forest-based anomaly detector identifies the onset of abnormal traffic states, and Dynamic Programming-based segmentation partitions the network into concentric impact zones (R1-R3) centered on the incident location. At the microscopic level, a Region Growing algorithm traces the spatial propagation path of congestion across adjacent links, while a Susceptible-Infected-Recovered (SIR) model quantifies the congestion onset, peak, and recovery timing for each link. Results demonstrate that incident-induced congestion propagates not randomly, but in a structured, hierarchical pattern, with recovery delays accumulating progressively in outer zones. The framework outperforms conventional static indicators by revealing where congestion originates, how it spreads, and which segments recover last.