Objective: The casualties and property damage resulted from crashes involving trucks are often more serious compared to crashes of other types of vehicles. Thus, this study was conducted to develop a model for studying factors affecting the crash severity of drivers in passenger car-truck crashes. Methods: A total of 28,605 valid accident data were extracted from Canada from 2007 to 2017. Using the police-reported variables, a classification and regression trees was used to identify significant predictors of passenger car-truck crashes. To handle the data imbalance issue, SMOTE and SMOTEENN algorithm were respectively applied to do the data preprocessing. The result shows that the SMOTEENN algorithm is more efficient than the SMOTE algorithm, in which the classification accuracy of the minority class increase from 51.4% to 73.1%. Results: Five significant predictors of driver-injury crash rate were identified, including: the direction of travel, the gender of car driver, roadway configuration, traffic control and whether to use safety device. Conclusions: The findings of this study contributed to design potentially useful policy initiatives as well as targeted safety promotion programs.

CICTP