Comparative Analysis on Multi-Classification Models in Traffic Accident Severity Prediction

The prediction of the severity of traffic accidents is regarded as a crucial measure to prevent major accidents in reality. This study compares and analyzes the effects of different multi-class models in predicting the severity of traffic accidents in order to clarify the most suitable model for accident prevention. Based on 84,833 road traffic accident data in Florida, USA, an ordered multi-class Logistics model (OLM), an ordered Probit model (OPM), and a support vector machine model (SVM) are established, and their performances are compared. The results show that the prediction of SVM on the training set and test set is better than that of OLM and OPM. On the training set, the prediction accuracy of SVM can reach 71.1%, while OLM and OPM are 70% and 69%, respectively. Besides, it is also noticed that all of the three models have multi-classification problems due to the large difference in the amount of different severity of the data, leading to the variation of the prediction accuracy of different levels of accidents. However, SVM still performs better in predicting level 1 and level 4 accidents. Therefore, compared with traditional mathematical statistical models, machine learning methods can better solve multi-classification problems.