High fuel consumption driving behavior identification and causal analysis based on LightGBM and SHAP

Abstract Accurate identification of high fuel consumption driving behaviors provides good theoretical support for eco-driving training. To gain a deeper understanding of contributing factors and their impacts on fuel consumption, this study acquired a driving data set based on a driving simulator test and employed the light gradient-boosting machine (LightGBM) algorithm to identify driving behaviors related to high fuel consumption and the SHapley Additive exPlanations (SHAP) algorithm for causal analysis. First, the vehicle kinematics and fuel consumption data were collected and preprocessed. Secondly, the LightGBM algorithm was employed to classify fuel consumption levels, including low, medium, and high. The results show that the LightGBM algorithm performs well in terms of precision, recall, and F1-score. Finally, the SHAP algorithm was used to interpret the influence of contributing factors on high fuel consumption from three perspectives, global interpretation, interaction interpretation, and individual interpretation. The SHAP algorithm can intuitively display the relationship between high fuel consumption and its contributing factors. Specifically, acceleration, speed, roll speed, pitch speed, and engine speed significantly increased the probability of high fuel consumption. This study proposed an efficient combined method for high fuel consumption identification and interpretation, which can reduce the occurrence of high fuel consumption driving behavior, thus achieving the purpose of eco-driving training.