A Machine Learning Framework for Analyzing the Causes of Traffic Accident-Induced Congestion

To explore the factors of traffic accident-induced congestion in depth and analyze the influence degree of each factor precisely, this study proposes a framework for analyzing the causes of interrelated traffic accidents and congestion. An evaluation of the influencing factors is conducted by comparing six algorithms, including logistic regression, support vector machines, decision trees, random forests, neural networks, and naive Bayes methods. The traffic accident data in the United States and the traffic collision data in California are selected and combined into a traffic accident-induced congestion data set with a total of 30,911 observations in California from 2017 to 2020. The influencing factors are identified and analyzed based on the feature importance. The results indicate that the random forest model yields the highest prediction accuracy, and the cause analysis framework is proposed based on this model. Theoretical suggestions for congestion mitigation and measures for preventing and resolving accident-related congestion are discussed.