A new traffic accident risk prediction method based on adaptive neural fuzzy inference system

To address the problem of traffic accident risk prediction, this paper conducts a study of objective factors by means of a fuzzy neural network system. The fuzzy neural network system includes fuzzy-based traffic accident risk prediction inference system which is used to derive traffic accident rules and adaptive neural network for traffic accidents factors learning which is adopted to self-learn and adjust the error. The interpretability of the fuzzy inference system combined with the powerful learning capability of neural networks allows fuzzy neural networks, not only predicts traffic accident risk but also translates the prediction process into interpretable and clear human language to facilitate further research on countermeasures. In this study, the data of 2017–2019 traffic accidents in Louis Anna state of United States were utilized as training and testing. Temperature, visibility, wind speed, and humidity were selected as independent variables and traffic accident severity as dependent variables for modeling. After experiments as well as comparisons, this method can achieve a prediction accuracy of about 78%, which is better than deep neural network and k-nearest-neighbor in machine learning algorithms, and the Gaussian-bell membership function applied is also the affiliation function with the best results.