InstanceEasyTL: An Improved Transfer-Learning Method for EEG-Based Cross-Subject Fatigue Detection.

Area of Science:

• Neuroscience
• Machine Learning
• Transportation Safety

Background:

• Electroencephalogram (EEG) signals are key indicators of driver fatigue.
• Cross-subject variability and limited data collection in driving scenarios pose challenges for EEG-based fatigue detection.
• Existing transfer learning models like EasyTL show promise but haven't been applied to cross-subject EEG fatigue analysis.

Purpose of the Study:

• To propose and evaluate an improved transfer learning classifier, InstanceEasyTL, for cross-subject EEG-based mental state detection, specifically driver fatigue.
• To assess the efficacy of InstanceEasyTL in reducing data requirements and enhancing performance compared to existing methods.

Main Methods:

• Development of InstanceEasyTL, an enhanced EasyTL-based classifier tailored for cross-subject EEG data.
• Comparative analysis of InstanceEasyTL against EasyTL, Support Vector Machine (SVM), Transfer Component Analysis (TCA), Geodesic Flow Kernel (GFK), and Domain-adversarial Neural Networks (DANN).

Main Results:

• InstanceEasyTL demonstrates superior performance in accuracy and robustness for cross-subject fatigue detection.
• The proposed InstanceEasyTL model requires significantly less EEG data for effective analysis.
• InstanceEasyTL outperforms established machine learning models in cross-subject EEG-based fatigue detection.

Conclusions:

• InstanceEasyTL offers a more efficient and effective solution for cross-subject EEG-based driver fatigue detection.
• The model's ability to perform well with less data addresses a critical limitation in real-world driving scenarios.
• InstanceEasyTL represents a significant advancement in applying transfer learning to complex neurophysiological signal analysis for safety applications.