Abstract
The occurrence of collisions between vehicles and bicycles leads to a considerable loss of human lives on a yearly basis, thus presenting a notable issue of concern for authorities responsible for ensuring the safety of traffic. Understanding the various factors that influence the severity of crashes involving vehicles and bicycles holds great importance. The application of machine learning ensembles involves the combination of multiple models in order to improve the accuracy of crash severity classification and prediction. However, the majority of ensemble frameworks are widely regarded as transparent systems, making it impractical to implement traffic safety judgments. This study presents three ensemble-imbalance learning (EIL) strategies, namely Self-Paced Ensemble, BalanceCascade and EasyEnsemble, in combination with advanced KTBoost, XGBoost, CatBoost, LGBM, AdaBoost and ET models as base estimators. This study aims at predicting the severity of vehicle-bicycle crashes. Based on data gathered by the Ningbo Public Security Bureau for the years 2020 and 2021, the EasyEnsemble method utilising XGBoost as the base estimator outperformed competing models. The Shapley Additive Explanations analysis was then utilised for interpreting the optimal model. The study found that the Weather Condition and Crash Location exhibited significant influence. The occurrence of inclement weather conditions, such as cloudiness or rainfall, in conjunction with the parking and the intersection zone, led to the occurrence of injuries.
Disclosure statement
No potential conflict of interest was reported by the authors.