| Compared with ordinary traffic crashes,the characteristics and occurrence mechanism of extraordinarily severe traffic crashes are different.This paper aims to study the distribution characteristics and influencing factors of extraordinarily severe traffic crashes.Based on the extraordinarily severe traffic crashes data from 2014 to 2018,the distribution characteristics of extraordinarily severe traffic crashes are analyzed from the aspects of drivers,vehicles,roads and environment,and the key influencing factors of the extraordinarily severe traffic crashes are revealed by the Logit model with random parameters and the machine learning model.The association rule technology is used to explore the correlation between the influencing factors of the extraordinarily severe traffic crashes,and the combination of the inducement factors of the extraordinarily severe traffic crashes is deeply excavated,so as to put forward more targeted safety measures.The main work and research results of this paper are as follows:(1)Analysis of characteristics of extraordinarily severe traffic crashes.The distribution characteristics of extraordinarily severe traffic crashes are analyzed from the aspects of drivers,vehicles,roads and environment,and the significant influencing factors of the number of fatalities are determined by ANOVA.The results show that the driver’s improper operation is the main cause of the extraordinarily severe traffic crashes.Trucks,secondary highways,national highways,straight road sections,southwest China,18:00-07:00,weekends,summer and sunny days are more likely to happen the extraordinarily severe traffic crashes,and speeding,overload driving,etc.have a significant impact on the number of fatalities.(2)Analysis of influencing factors of extraordinarily severe traffic crashes severity based on Logit model with random parameters.The extraordinarily severe traffic crashes are divided into three severity levels and the Logit model is built to analyze the key influencing factors of severity.The results show that,compared with the traditional Logit model,Logit model with random parameters has better goodness of fit.The variable of overspeed driving has random parameter characteristics.75.55% of the crashes involving overspeed driving will increase the number of deaths.The influence of overspeed,overloading,road snow or ice on the three severity levels crashes are different.(3)Analysis of influencing factors of extraordinarily severe traffic crashes severity based on machine learning.After balancing the extraordinarily severe traffic crashes data with SMOTE algorithm,the Boruta algorithm in feature selection is used to screen out the input variables of the model,and the decision tree C4.5,support vector machine and BP neural network classification models are established respectively to explore the main influencing factors of the severity of the extraordinarily severe traffic crashes.The results show that the overall performance of decision tree C4.5 model is the best.In order of importance,the main factors affecting the severity of extraordinarily severe traffic crashes are accident pattern,overspeed driving,road alignment and overloading driving.(4)Incentive analysis of extraordinarily severe traffic crashes based on association rules.The association rule Apriori algorithm is used to carry out association analysis on the influencing factors of extraordinarily severe traffic crashes.Through the analysis of the interaction mechanism of two factors,three factors and four factors,the rules such as frontal collision accidents of freight cars on national highways under bad weather conditions are more likely to occur due to wet road surface are excavated.Based on association rules,the typical accident inducement combinations of extraordinarily severe traffic crashes with different severity levels are mined,and then more targeted accident prevention measures are proposed. |
