Traffic Accident Severity Prediction Method Based on Decision Tree and Support Vector Machine

To research the injury patterns of road traffic accidents, classification tree, bagged-tree, random forest (RF), support vector machine (SVM), logistic regression, and stacking models are used in this paper to classify and predict traffic accident data set of a city in Shaanxi Province. The parameters of the six models are optimized, the prediction accuracy is assessed, and the key features influencing road traffic accidents are identified. The findings revealed that (1) classification tree and its bagging model performed well in sensitivity but not in specificity, whereas the support vector machine and logistic regression model did not; random forest and stacked model have high prediction accuracy; and (2) “vehicle safety condition,” “road type,” “type of roadside protection facilities,” “road physical isolation,” “lighting conditions,” and other variables all have a significant role in the accident’s severity. The study’s findings serve as a guide for traffic management divisions looking into traffic dangers.