Investigating determinants of injury severity in motorcycle-vehicle crashes using interpretable machine learning...

Due to inherent vulnerability, motorcyclists sustain more severe injuries in motorcycle-vehicle crashes. Although previous studies have analyzed diverse factors affecting injury severity of crashes involving a motorcyclist, few have been available to uncover how the factors and inherent interactions affect the motorcyclists' injury severities in motorcycle-vehicle crashes comprehensively. Thus, the study aims to investigate determinants of injury severity from multiple dimensions.

The study utilizes the machine learning models and the Random Parameters Logit model with Heterogeneity in Means (RPLM-HM) to explore the key factors contributing to injury severity sustained by motorcyclists. As an optimal machine learning model, CatBoost is used with the Shapley Additive exPlanations (SHAP) method to reveal the potential interaction between the crucial multidimensional factors.

Results indicate that 1) factors such as angle crash, elderly motorcyclists, high-speed limit, and risky behavior, are linked to a higher probability of severe injuries, 2) conversely, some features, including young motorcyclists, elderly drivers, wet road, and morning, are associated with reduced probability of severe injuries, and 3) there are interaction effects between the factors regarding the characteristics of motorcyclists, drivers and accident circumstances.

The findings could help put forward targeted countermeasures to better manage motorcyclists, drivers, and road facilities for improving the traffic safety of motorcyclists.