Evaluation of contributing factors for at-fault motorcycle rider crash severity in Sri Lanka using regression and machine learning techniques

Abstract

This study aims to investigate the contributions of roadway, environmental, vehicle, collision, and driver-related factors to the severity of at-fault motorcycle crashes using both Logistic Regression (LR) and machine learning (ML) approaches. Approximately 200,000 motorcycle-related crash records from 2008 to 2020, obtained from the Sri Lanka Accident Data Management System (SLADMS), were analysed. The study classified crashes into two categories: high-severity crashes, including fatal and grievous crashes, and low-severity crashes, including non-grievous and property damage only crashes. Five categories of contributing factors were considered: driver-related, roadway-related, vehicle-related, environmental, and collision-related factors. Six ML models, namely Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGB), K-Nearest Neighbour (KNN), Support Vector Classifier (SVC), and Artificial Neural Network (ANN) were compared with Logistic Regression using an 80/20 train-test split. Shapley Additive Explanations (SHAP) were employed to interpret the ML models. Among the evaluated models, the RF model demonstrated the best performance in predicting high-severity crashes, achieving a recall of 0.87, a precision of 0.60, and an F1-score of 0.71. SHAP analysis indicated that collision type (SHAP value = 0.085) was the most influential factor contributing to high-severity crashes, followed by the collided element (SHAP value = 0.068) and human pre-crash factors (SHAP value = 0.045). The findings suggest that explainable ML methods can be effectively used to assess crash severity and to facilitate more informed road safety decision-making.