Machine Learning-based Road Traffic Accident Severity Analysis And Prediction

In recent years,with the rapid progress of economy and society and the advancement of urbanization,the number of motor vehicles is also on the rise.However,followed by the increasing traffic safety pressure and the number of road traffic accidents remains high.Road traffic accidents not only cause a large number of casualties among the people of all countries,but also bring about a large number of medical,administrative costs,production and property losses.Therefore,how to scientifically prevent road traffic accidents has become the focus of traffic safety research.Many researchers use machine learning model prediction of road traffic accident severity,so as to explore the influence factors of road traffic accident severity,but complicated factors related to road traffic accidents,including roads,environment,vehicles,personnel and other relevant factors,while most of the accidents in road traffic accidents for minor accidents,data is not balance,these features increase the difficulty of the severity of road traffic accident forecast.In this paper,through the integration of multi-source traffic accident data files,all the factors are considered comprehensively.For the non-balance of the data,RUS,ROS,SMOTE and ADASYN were firstly used to sample the data,and three models of logistic regression,random forest and artificial neural network were established for the sampled training set.Then,the optimal model is obtained by comparing the three models.This model can not only provide suggestions for preventing serious road traffic accidents,but also provide reference for the traffic law enforcement departments to determine the severity of accidents quickly.Through the analysis,it is found that the traffic departments should strengthen the education and management of young and old drivers,strengthen the monitoring of motorcycles,and pay attention to the construction of road lighting.There are three innovative points in this paper.First,this paper considers the existence of multi-vehicle accidents and multi-casualty accidents,and constructs variables that can describe multi-vehicle accidents and multi-casualty accidents when considering variables.In addition,chi-square test and MV test are used in this paper to screen variables,especially MV test,as a relatively new test method,has not been widely used in variable screening.Finally,in this paper,the data of multi-classification problems are balanced,and the macro F1 value and MAUC comprehensive evaluation model are introduced.