Identifying relevant EEG channels for subject-independent emotion recognition using attention network layers

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is used for emotion recognition models.
• Subject-independent models are more practical but face challenges due to individual EEG variability.

Purpose of the Study:

• To identify consistently relevant EEG channels across individuals for emotion prediction.
• To enhance subject-independent emotion recognition models using deep learning and attention mechanisms.

Main Methods:

• Applied attention mechanism layers to EEG data from three independent datasets (SEED, SEED-IV, SEED-V).
• Focused on identifying EEG channels crucial for emotion prediction in a subject-independent manner.

Main Results:

• Achieved average accuracies of 79.3%, 69.5%, and 60.7% on the datasets.
• Identified EEG channels along the head circumference (e.g., Fp1, Fp2, F7, F8, T7, T8, P7, P8, O1, O2) as most critical.

Conclusions:

• Highlight the importance of specific EEG channels for subject-independent emotion recognition.
• Suggests that focusing on these channels can improve the practical utility of EEG-based emotion prediction systems.