Comparing features extractors in EEG-based cognitive fatigue detection of demanding computer tasks

Area of Science:

• Neuroscience
• Cognitive Science
• Biomedical Engineering

Background:

• Cognitive fatigue detection is crucial for monitoring performance during demanding tasks.
• Electroencephalography (EEG) is a key tool for analyzing brain activity.
• Power spectral density (PSD) is a common but potentially suboptimal feature extractor for EEG-based fatigue classification.

Purpose of the Study:

• To compare the effectiveness of different EEG feature extraction methods for classifying cognitive fatigue.
• To identify the optimal feature extraction technique for detecting fatigue during computer-based tasks.
• To evaluate autoregressive (AR) method against power spectral density (PSD), power spectral entropy (PSE), and wavelet methods.

Main Methods:

• EEG data was collected during cognitively demanding computer tasks.
• Feature extraction was performed using PSD, PSE, wavelet, and AR methods.
• A Bayesian neural network was employed as the classifier.
• Performance was evaluated using accuracy on AX-continuous performance test (AX-CPT), psychomotor vigilance test (PVT), and Stroop task.

Main Results:

• PSD and PSE methods yielded an average accuracy of approximately 60%.
• The wavelet method showed a slight improvement with an average accuracy of 61%.
• The AR method demonstrated superior performance, achieving accuracies of 75.95% (AX-CPT), 75.23% (PVT), and 76.02% (Stroop task) (p < 0.05).

Conclusions:

• The autoregressive (AR) method is a more effective feature extractor for EEG-based cognitive fatigue classification compared to PSD, PSE, and wavelet methods.
• Utilizing AR features can significantly enhance the accuracy of detecting cognitive fatigue during demanding tasks.
• This research provides valuable insights for developing more robust EEG-based fatigue monitoring systems.