Will the social vulnerability index affect school bus crashes?

Abstract

School bus crashes have become an increasingly important public concern due to its direct relationship with child transportation security and community well-being. Understanding the environmental and social factors can reduce school bus crashes, which is essential for exploring effective prevention strategies. Nevertheless, existing studies on school bus crashes often ignore spatial dependencies among communities and the combined effects of the built environment and social vulnerability factors. To address the aforementioned research gaps, this study firstly investigates the determinants of school bus crashes. We integrate the Connecticut school bus crash dataset with built environment variables and Social Vulnerability Index (SVI). Subsequently, Graph Convolution Network (GCN) is applied for geographical correlation in study. Moran's index of the residuals indicates that the GCN can capture geographic correlations. Finally, spatial cross validation results prove that SVI related to racial and ethnic minorities, housing and transportation play significant roles in crash occurrence. Other factors such as population migration and land use mixture also contribute to crashes. Variable thresholds and feature importance rankings are determined by SHapley Additive exPlanations (SHAP). Based on accident causation theory, this study enhances the understanding of how built environment and social vulnerability indicators jointly affect school bus crashes and offers insights for urban transport policy formulation.