A regression method for EEG-based cross-dataset fatigue detection.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Fatigue poses significant risks in jobs requiring sustained concentration.
• Existing electroencephalogram (EEG)-based fatigue detection models require extensive data for retraining on new datasets, which is impractical.
• Cross-dataset fatigue detection offers a solution but remains largely unstudied.

Purpose of the Study:

• To develop a novel EEG-based cross-dataset fatigue detection model.
• To address the limitations of retraining existing models for new datasets.
• To establish a robust method for fatigue detection applicable across diverse EEG datasets.

Main Methods:

• A two-step regression method involving pre-training and domain-specific adaptation.
• A pretext task during pre-training to differentiate between datasets.
• Domain adaptation using a shared subspace, Maximum Mean Discrepancy (MMD), attention mechanisms, and Gated Recurrent Units (GRU).

Main Results:

• Achieved an accuracy of 59.10% and a Root Mean Square Error (RMSE) of 0.27.
• Significantly outperformed state-of-the-art domain adaptation methods.
• Demonstrated high accuracy (66.21%) even with only 10% labeled samples.

Conclusions:

• The proposed cross-dataset fatigue detection model effectively generalizes across different EEG datasets.
• This approach reduces the need for extensive data and retraining, offering a practical solution.
• The study provides a valuable reference for future EEG-based deep learning research.