Performance Comparison of Machine Learning and Deep Learning While Classifying Driver’s Cognitive State

Abstract

Driver fatigue is a major cause of the road accidents that occur throughout the globe. It has been observed that among total number of accidents, 20% are contributed from driver fatigue. Acknowledging the existing data it is clear that a notification system for driver fatigue is of at most importance. Over the past a large number of strategies have been tested out and among them EEG based systems have shown to be the most accurate and reliable to estimate driver’s cognitive state. The direct relation of brain activity to EEG signal explains its high accuracy in a fatigue detection system. Current researches in machine learning as well as deep learning have shown a new perspective in EEG data analysis. This work proposed a highly accurate, EEG based driver fatigue classification system which can reduce the rate of fatigue related road accidents using machine learning and deep learning algorithms. The results showed that the relative power of theta, alpha, beta and delta showed significant correlation to driver fatigue. The selected features were trained and evaluated using 20 well established classifiers in the field of driver fatigue. Among all the classifiers tested, the Fine Tree, Subspace KNN, Fine Gaussian SVM, and Weighted KNN were performed to the highest accuracy levels. Different performance metrics are used for this work and Deep Autoencoder and KNN are identified as the best suitable Deep learning and Machine Learning Algorithms for driver fatigue prediction with an accuracy of 99.7% and 99.6 % respectively.