Relationship between Mesoscopic Vehicle Group Dynamics and Microscopic Driver Behavior: A Driving Style Perspective

This study investigates the coupling relationship between mesoscopic vehicle group dynamics and microscopic driving behavior complexity under heterogeneous driving styles, aiming to reveal how local traffic environments are associated with differential volatile behaviors across driving styles. Based on high-precision vehicle trajectory data from the Highway Drone (highD) dataset, local vehicle groups are first identified using a spatiotemporal similarity-based clustering method, and drivers are classified into aggressive, moderate, and conservative groups through spectral clustering. Sample entropy is then extracted from vehicle acceleration, intervehicle distance, and time headway sequences to characterize behavioral complexity and uncertainty in driver’s microscopic maneuvers, based on which a comprehensive behavioral entropy (CBE) index is further constructed using principal component analysis. To identify high-complexity driving behaviors across different driving styles, a random forest model is developed using mesoscopic vehicle group features, including group density, average speed, and speed and distance differences between the neighboring lanes. Shapley additive explanations (SHAP) is further employed to quantify the predictive contributions of key features and to identify underlying association patterns related to behavioral complexity. Results show that different driving styles exhibit significantly different responses to the same traffic environment. Aggressive drivers are more sensitive to vehicle group density and average speed, and interlane speed differences exhibit a strong association with their unstable behavior. In contrast, moderate and conservative drivers show pronounced nonlinear responses to intralane speed differences and spatial compression from adjacent lanes. These findings indicate that the mesoscopic traffic environment has differentiated effects on driving stability and decision-making fluctuations among different driving styles, providing empirical evidence for precise and proactive interventions targeting heterogeneous driving populations.