An Efficient Machine Learning-Based Emotional Valence Recognition Approach Towards Wearable EEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Emotion artificial intelligence (AI) is increasingly used in healthcare and education.
• Facial expressions and speech tone are unreliable for emotion recognition due to manipulation.
• Electroencephalography (EEG) offers a reliable and cost-effective alternative for detecting genuine emotions.

Purpose of the Study:

• To develop a subject-dependent emotional valence recognition method for emotion AI applications.
• To identify the most relevant features, frequency bands, and EEG time slots for accurate emotion detection.
• To create a computationally efficient and reproducible EEG-based emotion recognition system.

Main Methods:

• Utilized the DEAP dataset for analysis.
• Computed time and frequency features from Fp1 and Fp2 EEG channels.
• Performed binary and multiclass classification using selected features and the alpha frequency band.

Main Results:

• Achieved 97.42% accuracy in binary classification using the alpha band, outperforming existing methods.
• Attained 95.0% accuracy in multiclass classification.
• Feature computation and classification completed in under 0.1 seconds with reduced computational complexity.

Conclusions:

• The proposed subject-dependent EEG emotion recognition method is highly accurate and efficient.
• Minimal features and few channels (Fp1, Fp2) achieve state-of-the-art performance.
• The method's reliability and reproducibility make it suitable for wearable EEG devices in emotion AI.