Performance Evaluation of Major Classification Algorithms forAggressive Driving Detection using CAN-bus Data

Abstract

Detection of driver moods associated to driving style such as drowsy, distracted, vigilant, calm, or aggressive driving is one of the mainproblems of Advanced Driver Assistance Systems and it obviously plays vital role in the prevention of traffic accidents. The main goalof this study is to compare the performances of major Supervised Learning based Classification Algorithms (SLCAs) for aggressivedriving detection, which is one of the fundamental problems for understanding driver mood or driving style through CAN (Control AreaNetwork) bus sensor data. These algorithms utilize CAN-bus data acquired by OBDII (On-board Diagnostics) socket of the vehicle. Inour experiments, to get ground truth data, many trials referring to aggressive and calm driving have been conducted by different subjectdrivers and these sensor data have been labeled as “aggressive” and “calm”. Afterwards, these transformed into training data to assessperformances of SLCAs. As a result, the Naïve Bayes Classifier has been found to be more successful than the others.