Research on predicting the severity of highway traffic accidents based on BP neural network optimized by genetic algorithm

Traffic accidents prediction is always a hot topic in the field of road traffic safety. In recent years, the rapid development of artificial intelligence algorithms has brought new methods for accident prediction. This research focuses on artificial intelligence prediction methods and significant influencing factors to prevent road traffic accidents and reduce their severity. Based on the STATS 19 (2018-2019) accident dataset provided by UK Department of Transport and through literature review, 11 factors that relate to human, vehicle, road, and the environment were identified, and 10 of them were qualified to be the input variables after SPSS correlation analysis. Taking the severity of the accident as the output variable, the dataset was used to train three types of models: Back propagation (BP) neural network, Random Forest (RF) and BP neural network combined with genetic algorithm (GA-BP). On the basis of the GA-BP neural network model, this research utilized sensitivity analysis to identify important accident factors and used the model to analyze their impact features on the severity of accidents based on the real data. The results demonstrated that GA-BP neural network performs best in prediction efficiency, with an accuracy of 85.2% and an F1 score of 79.5%. In comparison, the prediction accuracy of BP neural network and RF was 83.5% and 66.0%, with an F1 score of 73.6% and 54.5%, respctively. Furthermore, during the training process, the GA-BP neural network outperformed the BP neural network in terms of convergence effect. Sensitivity analysis revealed that the most significant factors influencing accident severity were speed limit, road alignment and weather, while the least were the driver's gender and age. Predicted results and actual data suggested a positive correlation between the speed limit and the severity of accidents, which aligned with reality. Complex road alignments and adverse weather conditions tended to increase accident severity. This study aims to provide a new method for road accident prediction research and to offer a decision-making basis for improving the traffic safety.