Aggressive Driver Behavior Detection Using Multi-Label Classification

Abstract

Autonomous driving and advanced driver assistance systems aim to add comfort and safety to transportation. One major challenge facing advanced driver assistance systems is detecting aggressive driving. Aggressive driving behavior is a radical reason for fatal accidents. The driving environment is one compelling aspect affecting aggressive driving behavior. However, driving environment data are expensive and not easy to get. Thus, this work proposes a novel approach for aggressive driving detection that considers the driving environment by predicting it as a target class and considers the relationship between the driving behavior and the driving environment. Specifically, the proposed approach formulates the problem as a multi-label classification problem where the predicted classes are the driver behavior style and driving environment. We adopt several multi-label algorithms, including binary relevance, classifier chains, label powerset, and RAkEL. Moreover, we apply two classifiers: Random forest and Support vector machines. Furthermore, we investigate the impact of feature selection on classification performance. We performed the experiments on a real-world dataset. The accomplished results illustrate the superiority of the multi-label approach in aggressive driving behavior detection. In addition, feature selection significantly enhances the classification performance.