Research On Prediction Models Of Freeway Traffic Accidents Based On ZINB And Tobit Regression

Transportation brings great convince to people,however traffic accidents pose a great threat to the safety of people's life and property,especially on highways,once accidents happen,they extremely serious consequences will happen.By studying the traffic accident data of freeways,establishing a scientific accident prediction model and accurately predicting the number of traffic accidents,it can provide basis for freeway managers to formulate traffic safety improvement plans to improve traffic safety.The main content of this paper is to establish prediction models of freeway traffic accidents based on the ZINB and Tobit regression,and compare the models with the NB model,evaluate fitting effect and prediction abilityo each model.Eventually,the ZINB and Tobit regressions are deemed to be more suitable than NB regression for the traffic accident data regression,the prediction models of freeway traffic accidents based on ZINB and Tobit regression are better than the NB model.Firstly,the data of 4657 traffic accidents and related influencing factors were collected as research objects from Jingzhu,Yuegan and Kaiyang freeways in Guangdong province.The road sections are divided according to the homogeneity of road alignment design by using the indefinite length method,and 5,573 traffic accident prediction modeling samples are obtained.According to the traffic accident samples obtained,the characteristics of traffic accident data are analyzed.It is believed that the data is excessive dispersed and contains too many zero values,thus the traffic accident prediction models established based on ZINB and Tobit regression are reasonable.Secondly,based on system clustering and quartile method,the outliers in traffic accident data are eliminated.Based on the variance inflation factor method,the collinearity diagnosis of the explanatory variables of the models was carried out,and the explanatory variables with collinearity problems were processed.Then,the basic theories of the models were studied and compared,and the forms,principles and solving methods of NB,ZINB and Tobit models were mastered.The methods of model independent variable selection,significance test and regression effect evaluation were studied.Finally,variables of NB,ZINB and Tobit models were selected,and 16,24 and 16 independent variables were determined respectively.Based on the results of model variable selection,the modeling of NB,ZINB and Tobit models was completed by using the traffic accident samples.The fitting effect and prediction ability of NB,ZINB and Tobit models are analyzed and compared.The results show that,compared with NB model,ZINB and Tobit models can better fit traffic accident data and have better prediction ability,and the difference of fitting effect and prediction ability between ZINB and Tobit models is not obvious.