Driving Risk Assessment Based on Acceleration-Enabled G-G Diagrams and K-Means Clustering: A Naturalistic Driving Study

With the pervasive integration of sensors in driving recorders and smartphones, the acquisition of extensive driving data now enables the comprehensive evaluation of drivers’ day-to-day behaviors and to identify driving risk. This study proposed an innovative methodology for assessing driving risk, by integrating G-G diagrams and unsupervised machine learning algorithms to identify inappropriate driving patterns. In order to accomplish this, this study carried out a number of naturalistic driving tests in Wuhan, China, gathering vital parameters like speed, lateral acceleration, and longitudinal acceleration. By thoroughly analyzing these parameters, the K-Means clustering algorithm was employed to classify segments of driving behavior and encode the morphology of the G-G diagram to delineate the distinguishing features of individual driving patterns. The results reveal a notable correlation between the magnitudes of longitudinal and lateral acceleration, providing crucial insights for accurately assessing drivers’ styles. Furthermore, our investigation shows that when using identical acceleration and brake modes, driving at higher speeds is associated with a greater risk of unsafe driving behavior. The proposed framework not only advances the understanding of driving behavior patterns but also lays the foundation for the development of personalized driver-assistance systems, aimed at improving road safety and reducing the incidence of traffic accidents.