Investigating the impact of the built environment on traffic accidents using a Spatially Lagged X-Augmented Machine Learning framework

Investigating the relationship between the built environment (BE) and traffic accidents is crucial for improving urban traffic safety. To characterize the nonlinear spatial spillover effects and cross-regional interactions of BE features, this study proposes a Spatially Lagged X-Augmented Machine Learning (SLX-ML) framework, combined with SHapley Additive exPlanations (SHAP) for interpretation, while also examining the modifiable areal unit problem. The results show that the SLX-ML models perform best at the 1000 m scale, with SLX-LightGBM achieving the best performance. SHAP analysis indicates that BE features in both local and neighbouring areas have important effects on traffic accident occurrence,although local effects are stronger. Specifically, local POI mix, bus stop density, and building density exhibit clear threshold or saturation effects, whereas building density, building height, and transportation-service POI density in neighbouring areas demonstrate nonlinear spatial spillover effects. Furthermore, POI mix, bus stop density, and building density exhibit strong cross-regional interaction effects.