Prediction of Traffic Crash Severity Using Deep Neural Networks: A Comparative Study
Khaled Assi
Abstract
World health organization (WHO) reported that millions of people are killed and injured in road traffic crashes (RTC). The consequences of the increasing rate of traffic crashes include significant social and economic welfare loss. The severity of the RTC is an important element to investigate and address the welfare loss. Accurate prediction of RTC severity is beneficial to trauma centers as it generates crucial information that can be used to take the required actions that will help in reducing the aftermath of crashes. This study aims to evaluate the performance of deep neural network (DNN) in predicting the severity of traffic crash using attributes that can be identified quickly on crash sites. Moreover, the DNN model's performance was compared with that of the support vector machine (SVM) model, which is widely used for traffic crash severity prediction. Compared to SVM, it was found that DNN is superior in predicting RTC severity with prediction accuracy and F1 score of 95% and 93% respectively.