Driving style diversity in highway weaving areas: A drone-based analysis of population distribution patterns and...

Driving style heterogeneity significantly influences traffic safety and efficiency in highway weaving areas, yet how operational parameters systematically shape population-level behavioral patterns remains unclear. This study examines the relationships between weaving area operational characteristics and driving style population distribution patterns using drone-collected trajectory data from seven highway weaving areas, obtaining 12,707 complete vehicle trajectories. Six driving style features incorporating speed and acceleration metrics were extracted, and Gaussian Mixture Model clustering identified three distinct driving style types: aggressive (23.5%), moderate (58.9%), and conservative (17.6%). Population distribution characteristics were quantified using Shannon entropy, Gini coefficient, and dominance index, revealing significant variations across weaving areas with Shannon entropy values ranging from 0.376 to 0.703. Statistical analysis demonstrated that weaving complexity exhibited strong positive correlation with Shannon entropy (r = 0.909, p = 0.005), while headway time showed significant negative correlation (r = -0.760, p = 0.047). A critical threshold of 300m was identified for weaving length effects on population distributions. Machine learning-based driving style classification models incorporating 17-dimensional features achieved 95.4% identification accuracy, with lateral acceleration contributing the highest feature importance (35.5%) and operational parameters accounting for 31.4% of total feature significance. The findings reveal that operational parameters correlate with driving style distributions through hierarchical patterns where geometric parameters are associated with spatial constraints, traffic flow parameters provide dynamic regulation, and weaving task parameters create behavioral differentiation. This research contributes to understanding how infrastructure and traffic conditions systematically shape collective driver behavior patterns, offering new approaches for traffic management optimization in intelligent transportation environments.