A Data Mining Framework to Identify Important Factors of Fatigue and Drowsiness Accidents

Fatigue and drowsiness are the major factors contributing to accidents worldwide. According to statistics, 20 to 40 percent of traffic accidents in Iran are due to drivers’ fatigue. This study aims to identify the most important variables affecting the occurrence of fatigue and drowsiness accidents based on the classification and regression tree (CART) method. At first, 859, 378 police crash data of provinces Tehran, Fars, and Mazandaran during seven years (2011-2018) were segmented into homogeneous groups using the two-step clustering algorithm. Next, an oversampling technique is applied to deal with the crash data imbalance problem. Finally, the classification and regression tree combined with the boosting algorithm increases the accuracy of the models. The results of the classification tree showed that the main variables affecting the occurrence of fatigue and drowsiness accidents are: road type, time of day, road traffic direction, local land use, shoulder type, vehicle type, control type, and collision type. Moreover, the road type variable was the only significant factor in residential suburban areas of Mazandaran and Fars provinces. Also, the common variable in residential urban areas of all three provinces was the time of day. It was concluded that the combination of the CART algorithm with oversampling and boosting increased the accuracy of the models. Identifying influential factors in fatigue and drowsiness accidents in the three mentioned provinces could improve the engineering and executive interactions and appropriate educational programs.