Driver Behaviour Classification: A Research using OBD-II Data and Machine Learning

Abstract

Classification of driver behaviour has gained much attention due to its potential in a variety of applications, and On-Board Diagnostic (OBD) real-time data is often under-utilised. Hence, using On-board Diagnostic-II (OBD-II) data by categorising drivers based on their driving behaviour can be an efficient method. The objective of this study is to identify groups of drivers based on their driving styles using the collected OBD-II data. This study uses a Kaggle-obtained online dataset of OBD-II. The suggested model in this study analyses driving behaviour using both supervised and unsupervised methods. The relationship between all features and engine speed is analysed to select the optimal features, which include engine speed, vehicle speed, throttle position, and calculated engine load. Then, the proposed model makes use of the K-Means algorithm to create driving behaviour labels whether belong to safe or aggressive - validated by the safety score criteria. Different machine learning models including Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), AdaBoost (AB), Linear Combination (LC) and Weighted Linear Combination (WLC) are used, customised, and compared to get the most accurate prediction of driver behaviour. Experimental results indicate that the suggested driving behaviour analysis can reach an average rate of 98.72% accuracy using DT. However, implementing the ensemble method AB has improved the accuracy to 99.48%.