Research On Expressway Traffic Accident Frequency Prediction Method Based On Improved Negative Binomial Model

Traffic safety situation on expressway in China has been greatly improved compared with the past decade,but there is still a relative high accident rate,injury rate and fatality rate.Therefore,it is important to analyze the influencing factors of traffic accidents on expressway and explore the significant fators.Then,traffic accident frequency prediction model on expressway is developed,which will provide theoretical foundation and technical support for expressway management authority on safety management work.The data was derived from a certain section of Beijing-Zhuhai-Macao expressway(G4),which is located in the north of Guangdong province.Improved Negative binomial(NB)regression models were used to predict traffic accident frequency on expressway.Firstly,the influencing factors of traffic accidents on expressway were investigated from four aspects,namely driver,vehicle,road condition and environment.According to the actual data,16 independent variables were selected,and the definition and measurement of these independent variables were illustrated in detail.Secondly,four kinds of segmentation methods were put forward,which are traditional fixed-length method,slidable fixed-length method,grade consistent method and curve radius consistent method.16 independent variables were separetly calculated according to different segmentation methods.Thirdly,Non-linear negative binomial(NNB)regression model was used to develop traffic accident frequency prediction model based on past research and data structure.Three indicators,including log likelihood value,Akaike's Information Criterion(AIC)and Bayesian Infornation Criterion(BIC),were used to test the goodness-of-fit of NNB model.Similarly,three indicators,including mean absolute deviation(MAD),relative error(RE)and cumulative residual error(CER),were used to test the prediction accuracy of NNB model.To further judge the better model,the NNB model and the NB model were compared by the goodness-of-fit and the prediction accuracy.The elastic analysis was used to determine the influence degree of independent variables on the dependent variable.Finally,a combined models based on bivariate logit regression(BL)model and NB model was used to develop traffic accident frequency prediction model to predict two kinds of traffic accident frequency,which are the only property damage accident frequency and the casualty accident frequency,based on past research and data structure.Three indicators,including mean absolute deviation(MAD),mean squared error(MSE)and relative error(RE)were used to test the combined model.To further judge the better model,the combined model and two independent NB model were compared by the prediction accuracy.