Study On Expressway Traffic Accident Prediction Considering Data Heterogeneity

Expressway traffic accidents cost the lives of more than one million people all over the word every year.The biggest challenge for expressways is how to reduce the occurrence of traffic accidents and improve the road traffic safety level.Establish accident prediction model of expressway traffic accident can discovery the intrinsic mechanism of traffic accident and provide measures for accident prediction of expressway traffic accident.The occurrence of traffic accidents is a complex process.Uncontrollable temporal and unobservable environmental changes can lead to data heterogeneity of accidents during data collection,which will lead to biased results of model and incorrect inferences.Random effect model is an effective tool to solve data heterogeneity.This paper establishes a random effect negative binomial regression model to predict the frequency of expressway traffic accidents,which can solve the problem of data heterogeneity to some extent.This paper selects expressway in Washington as object,dividing segment in the way of variable length,selecting road alignment,operating speed,annual average daily traffic,road condition and sight condition as analysis factors,analyzing the influence of different factors on traffic accident and predicting the frequency of traffic accidents.Firstly,describe the heterogeneity concepts of different fields and the source of heterogeneity in factors which affect traffic accident.Secondly,the sensitivity analysis method is used to screen the divided road segments and a gradual selection method is used to screen variables to determine the independent variables and dependent variables.Furthermore,explore the influence of different factors on traffic accident rate.Finally,construct random effect negative binomial regression model to predict the frequency of traffic accidents under the condition of considering data heterogeneity and compare the result of random effect negative binomial regression model with negative binomial regression model which constructs under the condition of neglecting data heterogeneity.For comparison,the log likelihood test,AIC indicator and BIC indicator are used to test the fitting degree of the model and the relative tolerance,average absolute error,square sum error and root mean square error are used to test the accuracy of the model.The result shows that the fitness and accuracy of random effect negative binomial regression model are better than negative binomial regression model,which proves that prediction result is better when considering data heterogeneity and random effect negative binomial regression model can solve the problem of accident data heterogeneity to a certain extent.