Fatigue State Detection Through Multiple Machine Learning Classifiers Using EEG Signal

Abstract

Fatigued drivers can often cause long-distance accidents worldwide. Fatigue states are the primary cause of highway accidents. This study is conducted to provide a comprehensive and reliable fatigue state detection system to avoid accidents and make a good decision. Three machine learning algorithms were applied to seventy-six subjects' electroencephalogram (EEG) readings to test their performance. A preprocessing stage extracts relevant information before applying machine learning algorithms to the signal. Three analytical methods were employed in this study, specifically the Decision Tree, the K-Nearest Neighbors and the Random Forest. The study revealed that employing all the classifiers resulted in a satisfactory accuracy rate compared to existing state-of-the-art methods for detecting fatigue states. The classification accuracy using Decision Tree for four classes and two classes were achieved at 88.61% and 88.21% respectively, which can make this EEG-based technology a practical and dependable solution for real-time applications.

Keywords

• decision tree
• eeg signal
• fatigue detection
• k-nearest neighbor
• machine learning
• random forest