Toward Safer Highways: Data-Driven Approach to Detecting Aggressive Driving Using Connected Vehicle Technologies

Abstract

Aggressive driving behaviors such as tailgating and cutting off pose serious highway safety risks, especially for trucks. Timely detection of these behaviors can enable real-time interventions (e.g., automated driver warnings or vehicle safety system activation) to prevent crashes. This study presents a machine learning approach to detect tailgating and cut-off events using data from a high-fidelity driving simulator. Forty participants drove a truck in mild and heavy traffic scenarios within a connected vehicle (CV) environment, providing rich data for analysis. We fused four data sources—vehicle kinematics, CV-based metrics, road characteristics, and driver demographics—into five feature combinations to evaluate their predictive power. Four classification models (Artificial Neural Network, Support Vector Machine, Random Forest, and XGBoost) were trained on these feature sets. Performance evaluation across traffic scenarios shows that models leveraging CV data significantly outperform those using only traditional data, achieving high accuracy in identifying aggressive behaviors. Integrating CV features with conventional kinematic data substantially improved tailgating and cutting-off detection, underscoring the promise of CV technology for enhancing highway safety.