Background: Fatigue and drowsiness accidents are more likely to cause serious injuries and fatalities than other accidents. Statistics revealed that 20 to 40 percent of traffic accidents in Iran are due to drivers' fatigue. This study identified the most important factors affecting driver injuries in fatigue and drowsiness accidents.
Methods: The Classification and Regression Tree method (CART) was applied 11,392 drivers were in-volved in fatigue and drowsiness accidents in three provinces of Iran, over the 7 years from 2011-2018. A two-level target variable was used to increase the accuracy of the model. First, dataset in each of three provinces was classified into homogeneous clusters using a two-step clus-tering algorithm. Oversampling method was used for imbalanced accident severity datasets. Then, classification was improved by boosting method.
Results: The classification tree reveals that the month, time of day, collision type, and vehicle type were common factors. Also, driver's age was important in female drivers cluster; the geometry of the place and seat belt/helmet usage were important in urban roads cluster; and area type, road type, road direction, and vehicle factor were important in rural roads cluster. Also, the combination of the CART algorithm with oversampling and boosting increased the accuracy of the models.
Conclusions: The analysis results revealed motorcycles, lack of using a helmet or seat belt, curvy roads, roads with two-way undivided and one-way movement direction increased the injury and death of drivers. Collision with fixed object, run-off-road, overturning, falling, and defective vehicles increased the severity of accidents. Female drivers older than 44 years old have a higher probability of fatality. Identifying the factors affecting the severity of driver injuries in such accidents in each province could assist in determining engineering countermeasures and training educational programs to mitigate these crash severities.