Predicting Traffic Safety Using PSO-SVM Method and Back Propagation Neural Network

Road accidents and the resulting casualties are one of the current challenges of human societies that have imposed great economic costs on the economies of countries. Given the information on traffic safety in previous studies, it is very important to determine traffic safety planning in anticipation of an increase in traffic accidents. The neural network models used in this field have gaps. In this study, in order to solve the neural network problems after diffusion, a new method that combines particle group optimization and support machine (PSO - SVM) combined to be used to predict traffic safety. First, the factors affecting traffic safety and evaluation indicators are analyzed, then the traffic safety forecasting model is created by PSO - SVM according to the effective factors. Finally, traffic safety data from 1997 to 2018 are used to investigate the predictive ability of the proposed method. Experimental results show that traffic safety prediction by PSO - SVM is superior to post - diffusion neural network. MAPE values for predicting the number of events by PSO-SVM and post-diffusion neural network were 0.0281 and 0.0498, respectively. Models have more fluctuations than observation data, so more accurate models can be created to adjust these models. Due to the fact that the error in the data related to the number of injured is less than the data on the number of accidents and casualties, the cause can be related to the number of more data