A Geo-AI framework for deriving spatiotemporal determinants of crash fatalities

Road traffic accident fatalities remain a major public health issue worldwide. The recent increase in fatal crashes on intercity roads in Iran highlights the limitations of traditional safety analyses that focus on single risk factors and fail to capture the combined influence of driver behavior, road geometry, environmental conditions, and geospatial accessibility. This study proposes a data‑driven GeoAI framework for integrated analysis of fatal crashes using heterogeneous data sources, including traffic sensor observations, police accident reports, and road geometric and environmental information. A spatiotemporal contextual alignment process links each fatal crash to pre-crash conditions, road characteristics, environmental context, and accessibility features. MultiLayer Neural Network, Random Forest, XGBoost, and LightGBM are applied to the aligned dataset to identify key risk factors. Results show that behavioral indicators, road curvature, slope, shoulder condition, and proximity to medical and service facilities play a decisive role in fatal crash risk and hotspot formation.