Exploring Contributing Factors to Accident Severity Based on Random Forest Approach

Abstract

Traffic accidents carry severe consequences for both human life and property. Efficient traffic management necessitates not only a deep understanding of the underlying causes of these accidents but also the capacity to anticipate their severity. In this study, we delved into the factors that influence accident severity by analyzing data gathered from the Gotenba to Tokyo section of the Tomei Expressway in Japan during 2019. We applied a random forest model to a curated dataset of 701 cases to forecast traffic accident severity. Furthermore, a grid search was executed to pinpoint the optimal hyperparameters for this model. To evaluate the distinct impact of each factor on traffic accident severity, we utilized SHAP (SHapley Additive exPlanations) for visual representation. This methodology proved instrumental in highlighting high-risk variables and individuals. Significantly, our analysis pinpointed several findings, and one of these findings shown that accidents which transpired at the tail end of congestion zones exhibited a higher likelihood of severity. These robust findings pave the way for valuable insights that bolster expressway management.