Research On The Influencing Factors Of Urban Fatal Crash Types Based On Improved Mixed Logit Model

With the rapid development of motorization,China’s urban population and motor vehicle ownership have increased rapidly,which has led to the increasing number of the urban fatal crashes year by year.Road traffic fatalities grew at an average annual rate of5.6% from 2013 to 2018,so the situation of traffic safety(on urban roads)is becoming a critical issue.Based on the discrete choice model,analyzing the influencing factors of different fatal crash types,which is conducive to the in-depth understanding of the occurrence mechanism of various types of crash,so as to reduce the occurrence of fatal crash.However,in the application of traditional discrete choice models,unsupervised discretization algorithms are often used to discretize continuous variables,causing serious information loss of the discrete variables.Besides,this type of model cannot actively explore the interaction between variables,which is easy to cause wrong inference.To solve the above-mentioned problems,an improved mixed logit model is proposed to explore the influencing factors of urban fatal crash type,factors and the interaction effect between factors contributing to the occurrence of fatal crash types has analyzed the influence of various factors,which provide a reliable theoretical basis for the formulation of targeted traffic safety improvement measures.First,based on the fatal crash data in Shenzhen from 2014 to 2016,the crash types were divided into vehicle-to-vehicle,vehicle-pedestrian,and single-vehicle crash,and16 are independent variables are selected from the aspects of the driver,vehicle,road,environment.Secondly,given the serious information loss of the discrete independent variables caused by the traditional unsupervised discretization algorithm,the minimum description length principle(MDLP)in the supervised discrete algorithm was innovatively applied to the continuous independent variables in the fatal crash.The supervised discrete MDLP reduces the information loss of the discrete independent variables based on the principle of maximization of information gain,which is conducive to seeking better discrete point positions.Then,in order to overcome the adverse effects of ignoring the interaction between variables and leading to wrong inferences,the feature subset element algorithm based on association rule mining(FEAST)was used to mine the interaction between independent variables.Finally,based on the data processed by MDLP and FEAST,the mixed logit(ML)model was developed to identify the influence of various factors and the interaction between them on the fatal crash type,and the elastic analysis was performed to explore the influence of various factors on the probability of crash type.The results show that: The MDLP-FEAST-ML model proposed in this paper has a significantly better fit than the EWD(equal width discretization)-ML,MDLP-ML,and EWD-FEAST-ML model.Age of driver,vehicle type,location of road cross-section,road isolation form,road surface status,intersections and non-intersection locations,road alignment,lighting conditions,crash time,the interaction between driver’s age and vehicle type,and the interaction between driver’s age and lighting condition are significant related to to fatal crash types.The method proposed in this paper fully considers the information loss caused by the discretization of continuous variables and the influence of the interaction between variables on the type of crash,which is beneficial to reduce the occurrence of urban fatal crashes and provide a decision-making basis for improving road traffic safety.