Study On Factors Influencing The Severity Of Single Vehicle Crashes In Rural Highways Considering Data Heterogeneity

Rural highways form an important component of the highway network in China,and they also face prominent traffic safety issues.High fatality rates of single vehicle crashes on rural highways have attracted increasing attention.However,current researches focused on rural highway single vehicle crash are mainly concentrated in developed countries,and the conclusions of those researches can not be directly used as the basis for China to formulate rural highway traffic safety policy.In addition,crash data itself has a high degree of heterogeneity,which may cause deviations in the model estimation results and lead to erroneous inferences.Therefore,it is necessary to apply methods that can reflect the heterogeneity of crash data to explore the main factors affecting the severity of rural highway single vehicle crash in China,and to provide the theoretical basis for the formulation of traffic safety improvement strategies.First,this paper details the characteristics of rural highway single vehicle crashes in Anhui Province.And 28 factors were selected as independent variables from the aspects of driver,vehicle,crash related,road,traffic facilities and environment.Secondly,in order to reflect the heterogeneity of crash data,random parameter Logit model and latent class analysis-Logit model were utilized to explore the influencing factors of rural highway single vehicle crash severity in Anhui Province.Third,to overcome the limitations of latent class analysis in crash data clustering,a hybrid clustering-Logit method was proposed.Hybrid clustering method combining factor analysis and k-means was used to cluster rural highway single vehicle crash in Anhui Province,and established crash severity model for each cluster based on binary Logit model.Fourth,compare the above three methods in model goodness of fit,model prediction result and model estimate result analysis.Finally,based on the main influencing factors of the single vehicle crash severity,the road safety investigation of Sanhe Road was conducted to identify road safety status of rural highways in Anhui Province,and put forward targeted road safety improvement measures.Comparision results of three models show that hybrid clustering-Logit model proposed in this paper is superior to the random parameter Logit model and latent class analysis-Logit model in model goodness of fit,model prediction accuracy and reflecting the heterogeneity of the crash data.Hybrid clustering-Logit model clusters rural highway single vehicle crash data of Anhui Province into five clusters;there are significant differences in the factors affecting the severity of single vehicle crashes in each cluster: truck,curved road section and hill terrain are significant in several clusters,but they have different influence directions on the crash severity;gender,age,overspeed,lane type,pavement structure,pavement condition,whether the road traffic signal is complete,central separation facilities,roadside protection facilities and season are only have significant impacts on the severity of single vehicle crash in a certain cluster.