Resilient road safety modeling through spatially disaggregated explainable AI.

Abstract

Understanding spatial disparities in traffic accident severity is essential not only for improving transport safety but also for advancing sustainable and inclusive transportation systems. Road crashes represent a persistent public health and equity challenge, directly linked to the United Nations Sustainable Development Goals (SDG 3: Good Health and Well-Being and SDG 11: Sustainable Cities and Communities). This study proposes an explainable, spatially segmented machine learning framework to examine urban-rural heterogeneity in crash outcomes, using disaggregated accident data from Kent, UK (2022-2024). By treating urban and rural systems as separate analytical units, the study captures risk heterogeneity that conventional pooled approaches often obscure. Among five tested models, Random Forest achieved the best performance and was further interpreted using SHapley Additive exPlanations (SHAP) to uncover how key factors differ in influence across spatial contexts. The results reveal a behavioral risk profile dominating in urban areas and infrastructure-driven risks in rural environments. These findings highlight the need for context-sensitive, evidence-based interventions that ensure transport equity across regions. Additionally, the study contributes to sustainable governance frameworks, enabling spatially adaptive risk mitigation, inclusive policy design and the long-term resilience of transport infrastructures.