A decision-oriented risk prioritization framework for traffic crash management using explainable machine learning and economic impact analysis

Le résumé fourni par la source

Traffic crashes impose heavy human, economic, and societal costs. Machine learning has pushed crash severity prediction to high accuracy, yet most of this work stops at prediction and says little about what traffic safety managers should actually do. Here we propose a framework that treats severity-conditioned crash risk as a prioritization problem, combining probabilistic modeling, explainable interaction analysis, and economic impact assessment in one structure. Severity probabilities come from calibrated supervised models, with calibration checked through Brier Score and reliability diagrams. SHapley Additive exPlanations then expose the dominant risk factors and how they compound. Crash scenarios are finally ranked by an economic loss–based Pareto analysis to locate the hours carrying a disproportionate share of the conditional burden. Applied to 5,676 crashes in Batman Province (2013–2022), a small set of scenarios accounted for over half of the conditional loss, pointing to time-targeted, severity-stratified interventions, though exposure-adjusted validation is still needed.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.