Evaluating Advanced Driver Assistance System Effectiveness in Rear-End Near-Crash Events: Driving Risk Assessment Using a Field Operational Test

Abstract

Driving risk assessment plays a crucial role in identifying risk factors for near-crash events and enables proactive safety measures to prevent crashes. This study evaluated the effectiveness of the headway monitoring and warning feature of an advanced driver assistance system (ADAS) in assessing driving risk and identifying factors associated with rear-end near-crash events using a field operational test. The primary objectives of this study were to cluster driving risk levels and to identify the factors influencing the driving risk level. When drivers were exposed to ADAS, near-crash events decreased during the active phase compared with the stealth phase, indicating that ADAS has the potential to reduce near-crashes. We adapted the K-means clustering approach to classify near-crash events based on driving risk during the braking process, resulting in three risk groups: low, moderate, and high. Multiple machine learning (ML) techniques such as k-nearest neighbors, support vector machine, decision tree, multinomial logit model, random forest, and XGBoost were used to classify driving risk levels. Overall, the XGBoost model outperformed all other ML models when predicting near-crashes. The XGBoost model achieved an exceptionally high overall prediction accuracy of 96.4% across all three risk groups. Feature importance analysis for the XGBoost model revealed that seven features significantly influence the driving risk levels in near-crash events, namely, occupation, maximum deceleration, ADAS, average deceleration, velocity when braking, percentage reduction in kinetic energy, and velocity reduction. This model has the potential to identify and evaluate the driving risk level of near-crash events in real-world driving conditions, thereby improving overall safety, and future upgradation of ADAS systems.