Prediction of Pedestrian Chest Injury Severity Using BP Neural Network

Background: Chest is one of the three commonly injury parts for pedestrians in car-to-pedestrian crashes. Although many efforts and studies have been done on chest injuries, mathematical model relationship between chest injury risk factors and chest injury severity hasn't been widely identified. Objective: This paper is aimed to investigate and predict the severity of chest injuries caused by car-to-pedestrian crashes though establishing BP neural network prediction models. Method and Material: This study is carried out by conducting an in-depth investigation of car-to-pedestrian crashes from 2010 to 2014 in Chongqing and applying different training functions and hidden layer nodes in BP neural network training to determine the best prediction scheme. Results: The study results showed that the training function and the number of hidden layer nodes had great influence on the prediction accuracy of neural networks. The best prediction result for chest MAIS was obtained by using trainlm function with 11 hidden layer nodes. Conclusions: Models presented in the present study reflected the non-linear relationship between the severity of pedestrian chest injuries and the vehicle impact speed and pedestrian age. Further-more, study results fully testified the feasibility and effectiveness of predicting pedestrian chest injury severity by using BP neural networks.