Predicting Injury Severity of Work Zone Crashes Along Florida Freeways

Abstract

With increasing demand for capacity improvement, future highway construction, and the need for newer infrastructure, the United States is likely to experience more, longer duration, and longer stretches of work zones along its roadways. Work zones are often characterized by lower speeds, changes in traffic patterns, and sometimes narrower lanes, leading to safety compromises of workers, motorists, and other road users. As higher speed limits usually characterize freeways, the significant speed reductions along work zones create substantial shifts for drivers. The abrupt changes in speed and traffic patterns influence the occurrence of crashes as well as injury severity. This study used 2018 to 2022 Florida work zone data in conjunction with freeway crashes to determine the factors influencing the injury severity of crashes occurring along its freeway work zones. Six models, including the K-nearest neighbor, support vector machine, random forest, extreme gradient boosting, multinomial logistic regression, and ordinal logistic regression, were compared for prediction performance using various metrics. The random forest model was observed to be superior in its classification capacity and therefore used for the data analysis. Variable importance and partial dependency plots were used to interpret the model and understand the influence of roadway, environmental, temporal, and human factors associated with work zone crash injury severity. The findings from this study will be helpful to traffic engineers and other responsible agencies in freeway work zone planning to reduce crash injuries and improve roadway safety.