A novel and accurate EEG emotion classification model based on multiple attention local binary patterns.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalography (EEG) signals offer valuable insights into brain activity, emotions, and intentions.
• Machine learning models are increasingly used for EEG-based emotion classification.
• Existing methods face limitations in feature representation and subject-independent validation.

Purpose of the Study:

• To develop a novel EEG emotion dataset and a robust feature-extraction method.
• To create an advanced architecture for improved EEG emotion classification.
• To validate the model's performance in a subject-independent manner.

Main Methods:

• A new dataset of 14-channel EEG recordings from 22 participants was created, including arousal and valence labels.
• A multiple attention local binary pattern (MATLBP) function was developed for feature extraction, generating five feature vectors.
• A multi-level feature-engineering architecture incorporating feature selection and multi-classifier classification with iterative majority voting was implemented.

Main Results:

• The proposed framework achieved high accuracies on both self-collected (93.38% arousal, 88.64% valence) and DREAMER datasets (93.56% arousal, 97.22% valence, 86.73% dominance).
• Subject-independent validation demonstrated competitive performance compared to previous methods.
• The approach offered reduced computational complexity.

Conclusions:

• The MATLBP-based framework provides an accurate, computationally efficient, and subject-independent method for EEG emotion classification.
• Its performance suggests potential applications in clinical decision-support, neurofeedback, and assistive technologies.
• Further validation in larger, multicenter cohorts is recommended to confirm generalizability.