Drivers' Accident Frequency Distribution And Its Influencing Factors

The road traffic safety situation is grim at home and abroad.Human factors are major causes of accidents.At present,the study on the frequency of traffic accidents mostly are based on the "Community Area","Administrative Area" and "Intersection" to analyze the frequency of accidents and influencing factors.Few analyses of the accident frequency are based on drivers.Therefore,taking the car driver responsible for the accident as the research object,a zero-inflated model of the driver's accident frequency is constructed to analyze the distribution of driver's accident frequency and its influencing factors.This paper deals with non-equilibrium data based on SMOTE algorithm,and uses machine learning methods to deeply study the accident risk identification of drivers.The research can provide a scientific basis for using the historical data to provide drivers with accident warning.The main works are as follows:(1)Distribution of driver accident frequency.Through the cleaning and fusion processing of the original accident data,effective samples of 21,560 drivers were extracted.Using the method of basic statistical analysis,this paper analyzes the overall distribution of accident frequency and frequency distribution differences under different attribute characteristics.The results showed that gender,age,driving age,vehicle brand,number of violations by pilots in the previous year,type of violation,and type of first accident all had significant effects on the frequency of driver accidents.(2)Key influencing factors of driver accident frequency based on zero-inflated model.Using driver's characteristic variables and accident frequency data,a zero-inflated model of the driver's accident frequency is constructed to reveal the key impact factor of the accident frequency.The results show that the driver who had committed serious violations or whose first accident was an intentional offense is more likely to have another accident within a year.Drivers with long age,rigorous examination and many illegal driving behavior are tend to have high crash frequencies.(3)Driver accident risk identification based on unbalanced data.Based on the SMOTE algorithm,the accident frequency of non-equilibrium data is processed.Using Logistic model,SVM classification algorithm and CART classification tree,the importance of feature variables is sorted,and it is identified whether the driver will have multiple accidents in the whole year.The results show that the ranking of the importance of variables is the number and type of violations in the previous year,driving age,and the age of drivers,which demonstrated the rationality of accident warning for drivers based on historical violation data.For the recognition effect,the SVM method is the best,with an accuracy of 79%.