Crash Severity Classification Prediction and Factors Affecting Analysis of Highway Accidents

Abstract

Every day 3,700 people died in road crashes and many more suffer serious injuries. Road traffic collisions are not accidents; they are things that can be avoided. This paper’s objective was to develop a crash severity classifier based on previous road accident open data from the Ministry of Transport, Thailand. The confusion matrix was used as performance evaluation. The results found that the Gradient Boosting classifier outperforms other models. In addition, the identification of factors affecting crash severity is analyzed using the Shapley additive explanations (SHAP). The output revealed that features that contribute to a positive impact on more fatal accident severity are the number of trailers involved, tollway collisions, and overturn crashes. Whereas, the number of motorcycles associated, night-time collisions, and rear-end crashes gave a negative impact on the severity which led to lower injuries.