Recognition and interpretation of aggressive driving behavior for heavy-duty vehicles based on artificial neural network and SHAP

Abstract

Aggressive driving significantly impacts traffic safety, and heavy-duty vehicle drivers are more liable for causing serious crashes. This paper analyzes drivers' aggressive driving behavior from the vehicle type perspective and identifies the influencing factors of aggressive driving behavior through artificial neural network (ANN) and Shapley additive explanations (SHAPs). Using Kaggle's open-source aggressive driving data, we establish an ANN model to identify driving styles, where road conditions, environmental conditions, and vehicle parameters are independent variables and driving style is a dependent variable. The following measurements, including accuracy, recall, precision, and F1 score, are used to evaluate the model's performance, and the neural network got 85.33%, 82.32%, 84.16%, and 0.8308, respectively. To illustrate the influence of independent variables, the SHAP algorithm is used to analyze the model's feature importance. It was found that illumination and weather conditions influenced the model's performance along with the vehicle length. The number of lanes relates to driving style, and there were more aggressive driving behaviors on two-lane roads than on single-lane roads. Besides, heavy-duty vehicle drivers were more likely to drive aggressively in wet road conditions and indulge in aggressive driving behaviors at night. Particularly, drivers of heavy-duty vehicles were more likely to drive aggressively, provided that the vehicle in front was also a heavy-duty vehicle. These findings inform heavy-duty vehicle drivers to reduce aggressive driving behavior. The information is suitable for inclusion in driver education programs, thus improving traffic safety.