The Research Of Traffic Accident Duration Prediction Based On Machine Learning

Since twenty-first century,with the development of science,Great progress has been made in China`s national economy.At the same time,the number of vehicles and traffic accidents has increased rapidly.Subsequent problems caused by traffic accidents,such as traffic congestion,would lead more losses.Thus,the traffic control department hopes to make reasonable strategies to reduce the losses.But,they can not make reasonable strategies because they can not make scientific analyses about the traffic accidents.Traffic accients` duration is one important part of traffic accidents.If traffic duration could be predicted,the traffic control department can make reasonable strategies and people can make reasonable travel plans.So,prediction of the traffic duration attracts more and more attention.Road traffic accidents contain lots of information.Traffic duration is also influenced by kinds of parameters,and the relationships between these factors that affect the duration of the accidents are complex and uncertain.These factors would make it difficult to select the raw data and analyze the duration of traffic accidents.However,current researches are not able to simultaneously realize the prediction of the accidents` duration and the analyses of the relationships between impact factors.Simultaneously realzation of these two goals is the major content of this paper.In order to realize the prediction of the traffic accidents` duration and the analyses of the relationships between impact factors,Bayesian network-SVM has been proposed in this paper.By using the Bayesian network,the relationships between impact factors have been obtained,and the useless factors in the original data have been eliminated.By using the simplified data and the support vector machine algorithm(SVM),the prediction of the duration of the accidents is realized.Gridsearch(GS)algorithm,genetic algorithm(GA)and particle swarm optimization(PSO)algorithm have been selected to optimize the parameters of support vector machine.According to the prediction results,the advantages and disadvantages of these three optimization methods have been compared and analyzed.And the optimized parameters are obtained by the particle swarm optimization(PSO)algorithm.By comparing the prediction results predicted by the SVM model and Bayesian network-SVM combination model,this paper has proved the advantage of the Bayesian network-SVM model in the prediction speed and that the prediction accuracy of Bayesian network-SVM model would not decline.