Domain adaptive deep possibilistic clustering for EEG-based emotion recognition.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Emotion recognition using electroencephalogram (EEG) data faces challenges due to neural signal variability and limited labeled data.
• Traditional domain adaptation (DA) methods struggle with generalization, particularly when noise induces domain shifts, often exacerbated by techniques like Maximum Mean Discrepancy (MMD).

Purpose of the Study:

• To introduce a novel framework, Domain Adaptive Deep Possibilistic clustering (DADPc), for robust EEG-based emotion recognition.
• To address limitations of existing DA methods, specifically their sensitivity to noise-induced domain shifts and poor generalization.

Main Methods:

• Developed DADPc, integrating deep domain-invariant feature learning with possibilistic clustering.
• Reformulated MMD as a fuzzy entropy-regularized one-centroid clustering task.
• Incorporated adaptive weighted loss and memory bank strategies for improved pseudo-label reliability and cross-domain alignment.

Main Results:

• DADPc effectively mitigates noise-induced domain shifts while preserving feature discriminability.
• Experiments on SEED, SEED-IV, and DEAP datasets demonstrated superior performance in emotion recognition.
• Significant improvements in accuracy and generalization were observed across cross-subject and cross-session scenarios.

Conclusions:

• DADPc offers a robust solution for practical EEG-based emotion recognition.
• The framework advances cross-domain EEG analysis by combining deep learning with possibilistic clustering.
• This research enhances the state-of-the-art in generalized emotion recognition from EEG signals.