Research on Tunnel Traffic Accident Prediction Based on Random Forest Algorithm and its Performance Evaluation

Abstract

In this scholarly investigation, we delve into the analysis of the acquisition rate and prediction accuracy of a random parameter model for tunnel traffic accidents, leveraging machine learning algorithms. By meticulously scrutinizing historical traffic accident data, we have pinpointed traffic flow and weather conditions as the two principal factors that significantly impact tunnel traffic accidents. To ascertain the optimal parameters for our model, we have employed diverse machine learning techniques, encompassing linear regression, decision tree, and random forest. Upon rigorous comparison of the training set and the verification set, the random forest algorithm emerged as the most proficient in terms of prediction accuracy and capture rate. In the experiment, the random forest algorithm achieved an accuracy of 88% when predicting tunnel traffic accidents, and performed well in key indicators such as recall rate and F1 score. In the experiment, the random forest algorithm achieved an accuracy of 88% in predicting tunnel traffic accidents, a recall rate of 82%, and an F1 score of 85%. The accuracy rate for predicting minor accidents is as high as 95%, but the accuracy rate for predicting major and catastrophic accidents is only 30%. These results highlight the advantages of the model in predicting minor accidents, while also pointing out the room for improvement in predicting serious accidents. These results show that the random forest algorithm has significant advantages and potential in the field of tunnel traffic accident prediction. These noteworthy numerical results underscore the potency of the stochastic parameter model for tunnel traffic accidents, when grounded in machine learning algorithms. Such a model offers high prediction accuracy and capture rate, thereby providing effective early warning mechanisms and preventive measures for enhancing tunnel traffic safety.