Comparison of Numerical and Regressional Accuracy of RBF and FFBP Neural Networks in Prediction of Pedestrian Accidents

Abstract

Pedestrians have the lowest level of protection in traffic accidents, and as a result, the slightest contacts between pedestrians and vehicles lead to severe injuries or even fatality. Therefore, providing a suitable method able to accurately analyse the accidents of this vulnerable group will be very useful. To achieve the goal of this study, two types of RBF and FFBP neural networks were used to predict the pedestrian-vehicle accidents of suburban roads in the city of Amol. This analysis was done using two years of accident data for twelve suburban roads and eight effective parameters in traffic accidents for the roads under investigation. Results indicated that the RBF model showed less flexibility with low-level amounts of accident data and parameters and consequently, presented answers with low accuracy, while the FFBP model gained higher accuracies and showed high precisions in the prediction of pedestrian-vehicle accidents. KeywordsAccident; Pedestrian; Neural Network; RBF; FFBP