Research

This study examines the composition and distribution of driving behavior embedded in real-world vehicle trajectory datasets that are widely used for autonomous driving research, traffic safety analysis, and traffic operation studies. Using large-scale trajectory datasets collected under different road environments, including highD, exiD, and pNEUMA, the study analyzes longitudinal driving indicators such as speed, acceleration, deceleration, and jerk to identify dataset-level behavioral characteristics. PCA and K-means clustering are applied to derive driving behavior clusters for highway and urban road environments, and decision tree analysis is used to interpret the combinations of indicators that distinguish each cluster. The results show that driving behavior can be categorized according to the interaction between driving speed level and longitudinal control intensity, forming stable, moderate, and aggressive patterns whose proportions vary across datasets and road environments. These findings highlight that the driving context and observation conditions embedded in each dataset are directly reflected in the composition of driving behavior, providing foundational insight for driving behavior analysis and dataset interpretation using real-world trajectory data.