Analysis of Truck-Involved Work Zone Crash Fatalities in Florida

This paper presents the results of an analysis focusing on recognition of large truck-involved work zone crash patterns. Recognizing the limitations of logistic regression models that were commonly used in crash severity prediction, this study applied machine learning techniques including data resampling and decision tree/random forest models. Using a seven-year large truck involved work zone crash data in the state of Florida, random oversampling and systematic oversampling techniques were explored. Decision trees and random forest models were consequently built for the raw and resampled datasets. From a methodological perspective, results showed that a combination of oversampling with ensemble random forests technique can significantly improve model performance in predicting fatality crashes. Primary contributors included pedestrian involvement, lighting conditions, safety equipment, driver condition, driver age, and work zone locations. In view of fatality patterns, results showed that a combination of different factors can significantly increase the probability of a fatal outcome. Regarding pedestrian crashes, factors such as dark not lighted conditions, distracted truck drivers, airbag deployment, and driver’s age (young drivers outside city limits, senior drivers inside city limits) were highly fatal. For non-pedestrian crashes, the combination of front airbag deployment with any restraint system other than shoulder and belt was quite fatal. Also, abnormal driver condition increased the risk of a fatal outcome. Additionally, the presence of female drivers (in view of multiple vehicle crashes) highly decreased crash severity, probably due to their more careful driving manner compared to males. Interestingly, driver actions and maneuvers as well as roadway design and other physical environment features (i.e., number of lanes, median type, grade and alignment) did not show significant contribution to the model.