Interpretable Cross-Subject EEG-Based Emotion Recognition Using Channel-Wise Features

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based emotion recognition is crucial for brain-computer interfaces (BCI) and healthcare applications.
• Accurate cross-subject emotion recognition requires methods robust to individual EEG variability.
• Existing methods often struggle with subject-specific differences in EEG data.

Purpose of the Study:

• To propose a novel method for predicting cross-subject emotion from EEG data.
• To develop a representation robust to subject-specific variability in EEG.
• To enhance the accuracy of emotion recognition in BCI and healthcare.

Main Methods:

• A channel-wise feature is proposed to represent spatial connectivity between brain regions.
• Pearson correlation coefficients are used to calculate symmetric matrix elements, handling subject variability.
• A two-layer stacked long short-term memory (LSTM) network processes channel-wise features for temporal analysis and emotion modeling.

Main Results:

• The proposed method achieved state-of-the-art classification rates on benchmark datasets.
• Achieved 98.93% and 99.10% accuracy for two-class valence and arousal classification on the DEAP dataset.
• Reached 99.63% accuracy for three-class emotion classification on the SEED dataset.

Conclusions:

• The combination of channel-wise features and LSTM effectively predicts cross-subject emotions from EEG.
• The proposed method demonstrates significant improvements in EEG-based emotion recognition accuracy.
• This technique holds promise for advancing BCI and healthcare applications requiring reliable emotion detection.