EEG Data Augmentation for Emotion Recognition Using a Conditional Wasserstein GAN.

Area of Science:

• Neuroscience
• Machine Learning
• Artificial Intelligence

Background:

• Accurate emotion recognition from electroencephalography (EEG) signals is challenging due to limited data.
• Machine learning models struggle with data scarcity in EEG-based emotion recognition.

Purpose of the Study:

• To enhance EEG-based emotion recognition by augmenting limited datasets.
• To introduce a novel data augmentation framework using Conditional Wasserstein GAN (CWGAN).

Main Methods:

• Developed a Conditional Wasserstein GAN (CWGAN) framework for EEG data augmentation.
• Employed Wasserstein GAN with gradient penalty to generate realistic EEG data in differential entropy (DE) form.
• Utilized quality indicators to select high-fidelity generated data for supplementing the dataset.

Main Results:

• The CWGAN framework successfully generated realistic and high-quality EEG data.
• Augmenting datasets with CWGAN-generated data significantly improved emotion recognition model accuracies.
• Evaluated performance on SEED and DEAP public EEG datasets, showing substantial accuracy gains.

Conclusions:

• CWGAN is an effective method for EEG data augmentation in emotion recognition.
• The proposed framework addresses the challenge of data scarcity in EEG signal analysis.
• Enhanced EEG datasets lead to more accurate machine learning-based emotion classification.