Emotional State Classification from MUSIC-Based Features of Multichannel EEG Signals.

Area of Science:

• Medical Data Science
• Computational Neuroscience
• Signal Processing

Background:

• Electroencephalogram (EEG)-based emotion recognition is crucial for cognitive state disclosure but computationally intensive.
• Traditional methods using non-parametric models like Welch's Power Spectral Density (PSD) for EEG feature extraction are complex and time-consuming.

Purpose of the Study:

• To apply the parametric Multiple Signal Classification (MUSIC) model for EEG feature extraction in emotion recognition.
• To tune MUSIC model parameters for enhanced discriminative feature extraction from multichannel EEG signals.
• To identify and address dataset flaws that may have inflated previous classification accuracies.

Main Methods:

• Utilized the Multiple Signal Classification (MUSIC) model, a parametric frequency-spectrum-estimation technique.
• Extracted features from multichannel EEG signals using the MUSIC model on the SEED dataset.
• Classified three emotional states using MUSIC-extracted features and an artificial neural network.

Main Results:

• Achieved an average classification accuracy of 97% for three emotional states.
• Demonstrated a 95-96% optimization in run time compared to Welch's PSD method for feature extraction.
• Identified dataset limitations impacting prior high accuracy claims.

Conclusions:

• The MUSIC model offers a computationally efficient and accurate alternative for EEG-based emotion recognition.
• Parameter tuning of the MUSIC model is critical for effective EEG feature extraction.
• The study highlights the importance of dataset validation in reliable emotion recognition research.