Modeling the Severity Level of Truck Crashes Using Binary Logistic Regression Method

It is one of the main priorities of safety researchers to reduce the severity of crashes by predicting the severity of crash damage. Large trucks pose safety concerns and are responsible for a high percentage of fatal crashes, so studying their crashes can help determine factors that contribute to crash severity. This study examines factors affecting the severity of large truck crashes in California using the Highway Safety Information System (HSIS). Driver, road, crashes, and vehicle characteristics were categorized as predictive variables. In this study, binary logistic regression (BLR) is used to model the level of severity of accidents and evaluate the weight of various predicting variables on the severity of injuries. Typically, various classical econometric methods are used to model the severity of accidents. Following is a discussion of how the modeling can be evaluated and optimized. Based on the modeling results of this research, the first three variables that are more important than other variables include: weather in clear conditions compared to the weather in other cases such as snowy conditions (with an exponential coefficient of 1.146), AADT in two cases more than 250,000 vehicles per day (with an exponential coefficient of 1.341) and between 100,000 and 250,000 vehicles per day (with an exponential coefficient of 1.202) are less than 100,000 vehicles per day compared to AADT (all three mentioned variables were statistically significant and had positive regression coefficients). Despite the overfitting caused by binary logistic regression, this model performs well on the data set of this study (the difference between accuracy of test and training data is 0.1%).