Using Semi-Supervised Domain Adaptation to Enhance EEG-Based Cross-Task Mental Workload Classification Performance.

Area of Science:

• Cognitive Science
• Neuroscience
• Machine Learning

Background:

• Mental workload (MWL) assessment is crucial for preventing accidents and ensuring operator safety.
• Current MWL classification models struggle with cross-task generalization, limiting their practical use.
• A performance drop occurs when models trained on one task are applied to another.

Purpose of the Study:

• To develop a method for effective cross-task generalization in MWL classification.
• To address the challenge of applying MWL models across different tasks and subjects.
• To improve the robustness and applicability of MWL assessment tools.

Main Methods:

• Proposed a semi-supervised cross-task domain adaptation (SCDA) method.
• Utilized power spectral density (PSD) features for MWL recognition.
• Evaluated the method on MATB-II, n-back tasks, and the COG-BCI public dataset.

Main Results:

• SCDA achieved superior cross-task classification performance on both internal and public datasets.
• Accuracies reached 90.98% ± 9.36% (own data) and 96.61% ± 4.35% (COG-BCI).
• SCDA demonstrated high average accuracy in cross-subject scenarios (75.39% ± 9.56% own data, 90.98% ± 9.36% COG-BCI).

Conclusions:

• The semi-supervised transfer learning approach using PSD features is effective for cross-task MWL assessment.
• SCDA offers a feasible solution for generalizing MWL models across diverse tasks and subjects.
• This method enhances the potential for real-world applications of MWL assessment.