Resilient road safety modeling through spatially disaggregated explainable AI.

Ye H; Wu B; Yuan D; Yang X; Peng Q; Kou R · 2026 · PubMed Central

DOI: 10.1371/journal.pone.0344380

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the persistent challenge of road traffic accidents by examining spatial disparities in crash severity between urban and rural environments. Motivated by the stagnation of road casualty figures in the UK and the limitations of conventional pooled modeling approaches that obscure localized risk heterogeneity, the research aims to develop a resilient, spatially disaggregated framework for road safety. The authors argue that urban and rural systems possess distinct risk profiles—behavioral in urban areas and infrastructure-driven in rural settings—requiring context-sensitive interventions to achieve transport equity and align with United Nations Sustainable Development Goals. To investigate these disparities, the researchers utilized personal injury collision data from Kent, UK, spanning January 2022 to November 2024. Kent was selected for its diverse road typologies, ranging from dense urban centers to rural corridors. The dataset was split into urban (n = 13,616) and rural (n = 13,246) segments based on official spatial classifications. After preprocessing, which included handling missing values, encoding categorical variables, and addressing class imbalance using BorderlineSMOTE, five machine learning models were trained: Logistic Regression, Support Vector Classifier, Decision Tree, Random Forest, and Gradient Boosted Trees. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics via stratified five-fold cross-validation. The best-performing model was further interpreted using SHapley Additive exPlanations (SHAP) to identify context-specific determinants of crash severity. The results indicate that Random Forest achieved the best overall performance among the tested models. SHAP-based interpretation revealed distinct risk mechanisms across spatial contexts. In urban areas, crash severity was dominated by behavioral risk profiles, with male drivers significantly overrepresented in fatal crashes, suggesting a link to aggressive driving behaviors. Conversely, rural environments exhibited infrastructure-driven risks, characterized by higher travel speeds, fragmented infrastructure, and delayed emergency responses. Notably, fatal accidents in rural areas involved a significantly lower median driver age compared to urban areas, highlighting the vulnerability of younger drivers in rural settings. Descriptive analysis also showed that while most accidents occurred under favorable weather conditions due to higher exposure, rural accidents were more frequently associated with crossroads and higher speed limits, whereas urban accidents clustered around roundabouts and lower speed limits. The significance of this study lies in its methodological contribution of combining spatial segmentation with explainable AI to uncover hidden heterogeneity in traffic safety data. By moving beyond aggregated analysis, the research provides actionable, evidence-based insights for designing spatially adaptive safety strategies. The findings support the development of inclusive policy designs that address specific urban behavioral risks and rural infrastructure vulnerabilities, thereby enhancing the long-term resilience of transport systems and promoting equitable road safety governance.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• demographic disparities