Road Crash Injury Severity Prediction Using A Graphical Neural Network Framework

Abstract

Road traffic accidents remain one of the leading causes of fatalities and severe injuries worldwide, demanding
accurate and timely injury severity prediction to improve emergency response and resource allocation.
Traditional statistical and machine learning models often fail to capture the complex spatial, temporal, and
relational dependencies in crash data. This study proposes a Graph Neural Network (GNN) framework to
predict road crash injury severity by leveraging the interconnections between crash features, road networks,
and environmental factors. By modeling crash events as graph structures, the proposed system enables better
feature representation and improved prediction accuracy. The framework integrates heterogeneous data
sources—such as traffic conditions, weather, road topology, and vehicle attributes—to provide actionable
insights for transportation agencies and emergency services.