Fatigue Detection Based on Multiple Eeg Features Machine Learning Results: A Follow-Up Study

Abstract

Fatigue is a significant factor affecting performance and safety in various occupations. This study investigates the efficiency of using multiple electroencephalogram (EEG) features for fatigue detection through a machine learning model, specifically in a long-term, high-stress environment. 14 undergraduate participants were monitored over five days, with EEG data collected at multiple intervals. Multiple EEG features such as power spectrum, entropy, and complexity from eight channels were analyzed. The results showed significant correlations between EEG features and fatigue degree, with machine learning models effectively classifying between fatigue and alert states. The model results based on Eye-closed EEG features showed relatively higher accuracy and correlation with sleep duration. Tsallis entropy and Rényi entropy were the "best features", and FP1 was the "best channel" relatively. A generalized linear model (GLM) further validated the machine learning results, re-confirming the fatigue detection model's efficiency and potential application in sleep duration estimation.

Keywords: EEG fatigue detection, EEG features, Feature selection, Machine learning, generalized linear model (GLM)