Study On Comparison Of Influencing Factors Of Traffic Accidents Between Online Car-hailing And Private Car

With the increase of private car ownership and the growth of internet economy,online car hailing,as a new type of internet transportation mode joined by private car,is becoming more and more popular.Due to its advantages of timeliness,convenience and so on,the user experience of the online car hailing has been greatly improved,making it usheres in a wave of explosive growth,while attracting the majority of private car users to invest in the operation of online car hailing.This new operation mode can not only effectively reduce the waste of idle social resources,but also provide convenient conditions for traveler.However,it also brings many social problems,such as traffic accidents.The traffic accidents of network car hailing will not only cause difficulties in claims settlement,but also damage personal health,property loss and other issues.As the online car hailing is joined by private cars,there must be a certain connection between their traffic accidents.At present,the research on the influencing factors of private car traffic accidents is relatively mature,but there is less research on the accidents of the online car hailing.Therefore,this paper makes a comparative analysis on the influencing factors of the traffic accidents between the online car hailing and the private cars,in order to make suggestions on the conditions of the online car hailing and the private car traffic accidents as well as the registration of the private car as the online car hailing.Taking the network car accident and private car accident as dependent variables,logistic regression model,decision tree model,random forest model and xgboost model are used to model and analyze,and the effect of the model is discussed.For the problem of unbalanced data classification in the modeling process,the original data is divided into training set and test set according to 7:3.After that,the training set is processed by three methods: setting category weight,smote method and tgan method,and tested on the test set to compare the effect of each model.In the model evaluation index,AUC value,g-mean value and F1 value which can comprehensively reflect the classification effect of online car Hailing(positive category)and private car(negative category)are selected as the main evaluation indexes.Recall rate,which reflects the classification effect of network car hailing,is the secondary evaluation index,and refers to the classification accuracy and recall rate index.Finally,the optimal effect model is selected,and the improvement measures and suggestions to reduce the traffic accidents of network car hailing and private car are put forward.By comparing the model effects of four classification models in the state of data imbalance(without setting category weight)and considering the data imbalance treatment(setting category weight),the necessity of data imbalance treatment is obtained.Under the setting of category weight,the effects of four models are sorted as follows: XGBOOST > RF > Logistic >CART.Under the SMOTE method,there is little difference between RF model and CART model in comprehensive effect,which are better than XGBOOST model and Logistic model.Under TGAN method,the results of the model are: XGBOOST > RF > CART >Logistic.Under different data imbalance processing methods,the effect of each model is different.In Logistic model and XGBOOST model,the method of setting category weight is better than TGAN method,and better than SMOTE method.In CART model and RF model,TGAN method is better than the method of setting category weight,and better than smote method.Finally,the XGBOOST model with category weight is the best.Based on the analysis of the model effect and results,it is found that there are differences in the following aspects: driver’s vision,number of passengers,driver’s gender,vehicle production year,lighting status,road defects.Using the SHAP visualization model,the model results are analyzed in detail,and the specific differences between influencing factors of carhailing and private car traffic accidents are obtained,and suggestions for reducing car-hailing and private car traffic accidents are given.Measures to make recommendations for private car registration online car rental.