Research on traffic accident fatality prediction based on BP neural network

Abstract

Through the analysis of the influencing factors and correlation of traffic accidents, the main indexes affecting the death toll of traffic accidents are GDP, population, number of motor vehicle drivers, highway mileage, highway passenger turnover, highway freight volume and highway freight turnover. GM (1,1) and BP neural network are used to fit and train the traffic accident death toll from 1998 to 2017 respectively. The average error of GM (1,1) fitting and BP neural network training is 9.22% and 1.95% respectively, which shows that the training effect of BP neural network is better than that of GM (1,1). Using GM (1, 1) and BP neural network model to predict the number of traffic accident fatalities in 2018-2019 respectively, GM (1, 1) predicts that the number of traffic accident deaths from 2018 to 2019 is 52000 and 47000 and BP neural network predicts that the number of traffic accident deaths from 2018 to 2019 are both 60000. The average error of GM (1,1) and BP neural network prediction is 21.4% and 4.8%, respectively, indicating that the prediction result of BP neural network is more accurate. The prediction method and results provide reference for the management of transportation departments, and realize the transformation from traffic accidents to prevention.