A comparative study on traffic violation level prediction using different models

Abstract

This study analyzed severity levels of traffic violations, in order to study the level of traffic violations, to provide a reference for traffic offense management and traffic accident prevention. In this paper, the relationship between driver sex, age, years of driving, the type of vehicle, the ownership of the vehicle and the severity of the traffic offense were analyzed by using the 2015 traffic violation data from Guangzhou. The Bayesian network's model was used to predict the level of traffic violations. The cumulative logistics model and the neural network model under the training set were compared with the Bayesian networks model. As a result of the accuracy comparison, the Bayesian networks is about 70%, the cumulative logistic model is about 47%, the neural network model is about 51%. The results show that the Bayesian networks model can better predict the level of traffic violations.