Research Of Highway Traffic Accident Based On Bayesian Network

Highway,with high speed,enclosed type,and whole fly-over design as remarkable characteristics,creates various advantages to vehicle traveling.Rapid development of highway not only improves directly traffic status quo at home,but offers strong support to China economic construction.Nevertheless,there is a trend that increasing traffic accidents unavoidably go with fast growing of highway mileage.In spite of being gradually decreased in quantity,highway traffic accidents are still serious threat to people's life and property safety.Therefore,deep analysis for highway traffic accidents is of great significance to strengthen traffic safety on highway,thus reducing or avoiding traffic accidents.This paper,based on research on distribution regularities of highway traffic accidents,contrastive analysis to present specific accident data and overseas study achievements on distribution regularities of highway accidents,with help of statistical analysis method,researches distribution regularities of highways accidents in time,climate,accident type and vehicle type,thereby working out remarkable characteristics of China highway traffic accidents.Further,this paper makes analysis of relationship between traffic accidents and four factors operating in a highway traffic system,that is,people,vehicle,roadway and environment,and then,makes analytical study on accident severity with support vector machines(SVM)algorithm,the selection of which is based on accident information's specialty of high dimension and nonlinearity.Trough application of SVM algorithm on building disaggregated model,this paper explores the collected data of highway traffic accidents,differentiating accident severity according to binary classification methods,thereby establishing non-linear SVM model.Last,this paper divides the vehicle involved in the collected highway traffic accidents into bus and coach accident,Coach-truck accident and truck accident.This paper introduces five index during the process of model building,namely,accuracy?sensitivity?specificity?Harmonic Mean of Sensitivity and Specificity(HMSS),to make effect evaluation to Bayesian Network analysis,and meanwhile,make clustering effect evaluation by using Receiver Operating Characteristic(ROC)curve.Corresponding Bayesian Networks are finally built directing at the whole accident database and three clusters.Study on structure of Bayesian Network helps to work out dependency relationships between accident influential variables under different clusters.