Crash severity prediction using a series of artificial neural networks

The National Highway Traffic Safety Administration (NHTSA) reported that over thirty-five thousand people lost their lives on U.S. roadways due to crashes in 2015, alone. Despite the fact that numerous advances were made in the traffic safety prediction area, many roadways are still facing a high frequency of severe crashes. This, in turn, causes a devastating impact on society in terms of human losses, economic costs, and environmental damage. Previous research has shown the successful capability of Artificial Neural Networks (ANNs) in predictive applications in a variety of environments. This research aims to identify/predict influential factors contributing to crash severity by carrying out a series of ANN analyses. Identifying the influential factors can aid in reducing both severity and frequency of crashes on the roadways. Eleven years' worth of crash data (2003-2013) from the State of Wyoming roadways is utilized in this effort to predict these factors. K-fold cross validation was used to validate the results.