Using robust principal component analysis to alleviate day-to-day variability in EEG based emotion classification

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Day-to-day variability in electroencephalography (EEG) data presents a significant challenge for accurate emotion classification.
• Emotion-irrelevant EEG perturbations can obscure subtle, emotion-related neural dynamics.

Purpose of the Study:

• To investigate the efficacy of Robust Principal Component Analysis (RPCA) in mitigating inter-day variability in EEG.
• To determine if RPCA-purified EEG data enhances emotion classification performance.

Main Methods:

• Employed RPCA to decompose EEG signals, separating emotion-relevant dynamics from background noise.
• Analyzed a five-day EEG dataset from 12 subjects.
• Compared classification performance using raw EEG features versus RPCA-purified EEG features.

Main Results:

• Empirical results validated the hypothesis that background perturbations cause day-to-day variability and mask emotion signals.
• RPCA successfully isolated sparse, emotion-related EEG dynamics from confounding daily variations.
• Utilizing RPCA-purified EEG data across multiple days led to a steady improvement in emotion classification accuracy.

Conclusions:

• RPCA is a feasible method for alleviating ecological inter-day variability in EEG data.
• Integrating RPCA into machine learning frameworks can enhance the robustness of affective brain-computer interfaces (ABCIs).
• Longitudinal data analysis combined with RPCA shows promise for personalized and reliable emotion recognition systems.