Study On The Relationship Between Urban Road Traffic Accident Duration Prediction And Accident Influencing Factors

The traffic condition of urban roads affects the development of a city,so it is especially important to ensure the smooth flow of urban road traffic if an accident can be solved quickly.Occasional congestion caused by traffic accidents is often seen in real life,and it is important to provide timely information to the vehicles arriving at the road section so that the vehicle owners can have a more accurate prediction of the accident duration and thus make a choice of the travel path.In this paper,we collected traffic accident data of Beijing’s fourth ring road for a whole year,totaling 5388 accidents,and used random forest classification algorithm,survival analysis and multinomial logit model to rank the importance of factors influencing accident duration and to predict the accident duration.Firstly,the raw data were organized and graphs were generated using the data to visually explore the relationship between each influencing factor and duration.Secondly,the random forest classification algorithm was used to study the accident duration as a classification problem and rank the importance of the influencing factors,and the accuracy of the final classification result was 90.3%.The K-M method in survival analysis was then used to explore the relationship between the influencing factors ranked in the front of the random forest obtained and the accident duration.Then Cox regression was used to investigate the importance ranking of the influencing factors of accident duration and to compare the results with the importance ranking obtained from random forest.Finally,a multinomial logit model is used to predict the total accident duration,accident response time,and accident clearance time,and the accident response time and clearance time predictions are aggregated to obtain the aggregated accident total duration prediction results,which are compared with the directly obtained total duration prediction results.The accuracy of the directly obtained total duration prediction results is 84.4%,the accuracy of the response time prediction results is 88.5%,the accuracy of the clearance time prediction results is 87.2%,and the accuracy of the aggregated total accident duration prediction results is 87.85%,slight improvement in accuracy over the directly obtained prediction results.The innovation of this paper includes the comparative analysis of feature importance ranking by two methods(machine learning and statistical methods),which is more convincing than the importance ranking results obtained by traditional single method.The prediction of total incident duration is compared with the prediction of total incident response time and clearance time,so as to determine which prediction is more accurate and to provide an additional prediction method for incident duration.