Possibilistic Clustering-Promoting Semi-Supervised Learning for EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) are advancing emotion recognition.
• Graph-based semi-supervised learning (GSSL) shows promise but struggles with noisy electroencephalogram (EEG) data.
• Individual EEG patterns can introduce noise and outliers, challenging existing GSSL methods.

Purpose of the Study:

• To develop a robust and reliable method for EEG-based emotion recognition.
• To address the limitations of existing GSSL methods in handling noisy and outlier EEG data.
• To improve the accuracy of emotion recognition in brain-computer interfaces.

Main Methods:

• Introduced a Possibilistic Clustering-Promoting semi-supervised learning method.
• Constrained instance label membership to local weighted means for enhanced reliability.
• Incorporated a fuzzy entropy regularization term to improve robustness against noise and outliers.

Main Results:

• The proposed method demonstrated improved reliability and robustness in EEG-based emotion recognition.
• Experiments on DEAP, SEED, and SEED-IV datasets validated the effectiveness of the approach.
• Enhanced generalization ability of the membership function through increased sample discrimination information.

Conclusions:

• The novel Possibilistic Clustering-Promoting method significantly enhances EEG-based emotion recognition.
• The technique offers a more robust solution for real-world BCI applications dealing with noisy data.
• This work contributes to more accurate and dependable emotion recognition systems.