Probabilistic Versus Non-Probabilistic Machine Learning Approaches for Estimating the Severity of Crashes in Construction Work Zones

Roadway work zones often present hazardous conditions for drivers, pedestrians, and construction workers. Understanding the factors contributing to work zone crashes and their severity can assist the departments of transportation (DOTs) in safety management and planning. Accordingly, this study investigated the effect of freeway work zones on the severity of crashes by employing machine learning algorithms. The model was developed using crash data and construction zone information from Utah between 2017 and 2021. This study compares the performance of probabilistic and non-probabilistic machine learning models in predicting work zone crash severity using the KABCO severity scale. All three models achieved promising accuracy levels, with Extreme Gradient Boosting (XGB) achieving the highest accuracy at 86%, followed by Gaussian Naïve Bayes (GNB) with an accuracy of 76%, and complement Naïve Bayes (CNB) with an accuracy of 74%. The findings of this study offer valuable insights for the safety management plans of DOTs.