Driving Style Analysis of Car Drivers Using Real-Time Driving Data – An Unsupervised Approach

Abstract

This study aims to explore the range of driving patterns exhibited by passenger car drivers. The driving data is collected for 42 drivers using instrumented vehicles, in a naturalistic environment. An algorithm is developed for data segmentation and total 7548 acceleration events, 6156 braking events and the corresponding driving performance features are extracted. The similar patterns of events are identified using K-means clustering technique in combination with PCA. Total four patterns of acceleration and braking behavior each, characterized by the level of acceleration/braking behavior, speed behavior and the event duration are identified. The results show that, each driving pattern represents a unique driving behavior and each driver is found to exhibit different driving patterns. No driver constantly shows a single driving pattern and the proportion of patterns are changing among individuals showing inter-driver behavioral variability. The insights from the study aids in developing driver specific safety models for driver assistance.