Mental State Classification Using Multi-Graph Features.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Extracting high-level mental states like stress and cognitive load from electroencephalogram (EEG) data is crucial for various applications.
• Traditional methods often rely on band power features, which may not capture the full complexity of brain activity.

Purpose of the Study:

• To propose and evaluate a novel feature extraction method for multi-channel EEG data.
• To assess the effectiveness of these new features for classifying high-level mental states.

Main Methods:

• Utilized spectral-based multi-graph tools applied to the time series of statistical dependence structures (e.g., correlation) among EEG sensors.
• Compared classification performance using proposed features against traditional band power features on two independent datasets (>= 30 participants each).

Main Results:

• The proposed spectral-based multi-graph features provide complementary predictive information to traditional band power features.
• Classification performance was enhanced when combining both feature sets.
• Identified specific EEG channels and channel pairs with significant predictive importance, aligning with neuroscientific understanding.

Conclusions:

• The developed spectral-based multi-graph feature extraction method offers a valuable advancement for EEG-based mental state inference.
• This approach enhances the classification of complex mental states by capturing intricate sensor dependencies.
• The findings support the neuroscientific validity of the identified important brain regions for cognitive state monitoring.