Fatigue Detection in SSVEP-Based BCIs Using Biomarkers: A Comparative Study

In a BCI system, prolonged command execution can lead to mental fatigue, making the user dissatisfied and decreasing the system's effectiveness. Utilizing functional indicators to determine the level of fatigue is the first step in reducing its negative consequences. This study classified alert and fatigue states using a combination of fatigue indices for better classification performance. First,
$\boldsymbol{\alpha}, \boldsymbol{\beta}, \boldsymbol{\theta}$
frequency bands, SNR, and MSE were extracted from long-term and continuously recorded signals. The significant increases in
$\boldsymbol{\alpha}, \boldsymbol{\beta}, \boldsymbol{\theta}$
, as well as the decrease in SNR and MSE, are observed to be associated with the increasing fatigue level. Then, all of the fatigue indices were classified with SVM, KNN, Decision Tree, Ensemble, and Logistic Regression.
$\boldsymbol{\alpha}$
frequency band and MSE are the most appropriate fatigue indices because in the fatigue state the ability to process complex information decreases and the mental effort to maintain vigilance level increases. The main results show that a combination of fatigue indices can improve the accuracy of classification between alert and fatigue conditions to 88.5%.