Date of Graduation
Statler College of Engineering and Mineral Resources
Civil and Environmental Engineering
David R. Martinelli.
This research focuses on using artificial neural networks to classify the severity levels of crashes involving guardrails, and to subsequently identify the most significant variables explaining severity in such crashes. Most of the existing research in analyzing guardrail crashes employs statistical analysis to measure severity of crashes and, unfortunately, does not incorporate much information about the factors that affect the severity concerning guardrail crashes. In the mean time, artificial neural networks have been utilized in different areas of transportation to solve engineering problems because of their ability to model non-linearity, and flexibility with large complex data sets. Data for this research were obtained from the Highway Safety Information System and were divided into two groups, the first group included roadway characteristics including guardrail/environment as input, and severity was output. The results showed that light condition, road surface condition, end and type of the guardrail significantly affect severity levels. The second group included vehicle factors and human factors as input and crash severity was output. The resulting classification was significantly affected by the driver age and vehicle impact. Merging all factors in one model resulted in the best classification of different levels of severity (above 93% in testing classification for different class of severity) and MSE = 0.027089 in cross validation. The results have demonstrated that the Neural Networks are an effective tool to classify severity levels in crashes with guardrail if appropriate input data is available.
Shoukry, Fouad N., "Artificial neural network in classification of severity levels in crashes with guardrail" (2005). Graduate Theses, Dissertations, and Problem Reports. 1647.