Drivers' Rear-end Conflict Risk Avoidance Behaviors Prediction Based on Natural Driving Data

Abstract

Traffic conflict prediction is one of the important principles in Advanced Driving Assistance System (ADAS). Previous research mainly focuses on calculating the risk of conflicts with surrogate safety measures, while the prediction of driver's rear-end risk avoidance behavior after conflict is less considered. Therefore, this study intended to predict rear-end conflict risk avoidance behavior, for it has significant impact on the crash likelihood and would determine whether the conflict will end into a crash. The study identified two risk avoidance behaviors within conflicts, i.e., the deceleration degree and duration. The K-means algorithm was employed to categorize these two behaviors into several patterns. Subsequently, the contributing factors were selected through significant analysis and multicollinearity test, and two logistic regression models were developed for prediction. Model coefficients indicate that factors related to on risk avoidance behavior. Prior risk perception before conflicts and stronger traffic fluctuation will lead to a higher deceleration degree and a longer deceleration duration. Moreover, if drivers are aggressive in speeding up, a violent deceleration and longer brake duration can be expected, if drivers maintain a wide margin with high concentration, the brake time will also be longer. The area under curve (AUC) of the prediction model is 0.84 and 0.77 for the deceleration degree and duration, respectively. The findings of this study would help researchers predicting the likelihood of a conflict evolving into a crash, thus facilitating practitioners to develop a more personalized ADAS. IEEE