Towards a versatile mental workload modeling using neurometric indices.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Accurate mental workload (MWL) modeling is crucial for understanding cognitive states.
• High-dimensional electroencephalography (EEG) data presents challenges for effective MWL classification.
• Feature selection is vital for interpreting EEG data and improving classification model accuracy.

Purpose of the Study:

• To propose an optimized feature selection technique for classifying MWL at three distinct levels.
• To enhance the accuracy of MWL classification using EEG data.
• To identify robust features from EEG data that correlate with cognitive load.

Main Methods:

• Utilized Recursive Feature Elimination (RFE) for feature selection from EEG data.
• Examined brain signals from thirteen healthy subjects performing a visual task.
• Employed Support Vector Machine (SVM) for classification of MWL levels.

Main Results:

• The proposed RFE technique achieved an overall classification accuracy of 0.913, an improvement from 0.791.
• Significantly demonstrated variations in selected feature mean values across three MWL levels (p<0.05).
• Outperformed other feature selection techniques in classifying MWL.

Conclusions:

• The developed RFE-based model provides an effective method for quantifying MWL levels from EEG data.
• This approach has broad applicability in fields such as neuroergonomics, assistive device development, and cognitive assessment.
• The findings support the use of optimized feature selection for robust cognitive state monitoring.