Risk Modeling And Characteristics Analysis Of Urban Multi-mode Traffic Accident

Traffic accidents are mostly related to the driver.In order to improve urban traffic safety,source governance of driver has become one of the current problems in urban traffic management.On the one hand,traffic environment and driver behaviors in China are greatly different from those of Western countries.Thus,it is not suitable to copy foreign traffic management methods directly.On the other hand,most of the current research on driver risk modeling is based on traditional historical accident data,which will limit the recognition of accident risk mechanisms and the precision of high-risk driver identification.In terms of the above background,this study will be executed from the following three aspects on the three main modes of transportation in the city(car,truck,e-bike):(1)At first,this paper will make full use of multi-source traffic data to expand the analysis sample.(2)Then,a complete set of machine learning models will be introduced to identify high-risk drivers.(3)Finally,different model explanation methods will be adopted to analyze the influence of different features on different modes.The specific research content is as follows:Firstly,the risk features will be extracted based on multi-source data fusion.In this chapter,the research sample size will be expanded vertically through matching of traffic accident data and traffic violation data,in which the demographic features,the vehicle features,the road features,and environmental features are contained.Then,the shortest path-based path reconstruction method is used to extract exposure features of different drivers through data fusion of ALPR data and historical traffic data,which can horizontally expand the dimension of the analyzing sample.Through the integration of multi-traffic data,the quality of analyzing sample can be enhanced,which can improve the model performance effectively in the following study.Secondly,constructing ensemble machine learning models to identify high-risk drivers.Considering the harm of traffic accident on urban traffic system,the study will define drivers who have had a responsible accident as high-risk drivers,and drivers who have not had accidents as non-high-risk drivers.Based on multiple factors such as people,vehicles,roads,environment,illegality,etc.,a set of sample balancing techniques and machine learning models are adopted to model the risks of the drivers of different modes.According to the model performance,it is found that the Balanced Bagging resampling method can better improve the data quality and improve the accuracy of the model.As for the model,the GBDT model can better fit the non-linear relationship between the safety features and the traffic accident.Finally,the accident risk features of different drivers are discussed.This chapter will be conducted based on the best identification models of different travel modes.The method of Feature Importance and Partial Dependence Plot(PDP)are used to analyze the importance of features to the model and the impact of features on driver accident risk.In this chapter,the relationship between risk features and driver risk can be deeply analyzed with different travel modes.The final results of this paper can provide theoretical support for the traffic police department to deeply understand the differences of accident risk among different traffic modes in urban traffic system,which can improve the level of urban road traffic safety effectively.