Freeway traffic accident severity prediction based on multi-dimensional and multi-layer Bayesian network

In order to explore the influence of the characteristic factors of freeway traffic accident on the severity of the accident, this study established the freeway traffic accident severity prediction model influenced by road and environmental factors, explored the influence of road factors and environmental factors on the severity of the accident. Firstly, the XGBoost model was established and SHAP value was introduced to explain the XGBoost model, and the importance ranking of the influence degree of each feature on the target variable was obtained and the global influence of each feature on the target variable is visualized. Then, according to the selected variables and their values, the freeway traffic accident severity prediction model based on Bayesian network was constructed. Finally, based on the data of freeway accidents in Hebei Province, a case study was conducted to predict the severity of accidents by road and environmental factors, and then the severity of accidents under the interaction of different factors was predicted by using the constructed prediction model. The results show that the learning accuracy and prediction accuracy of the model are verified, and the performance of the prediction model is good. According to the analysis results, the corresponding safety management measures can be put forward to reduce the severity of the accident.