Cross-Day Data Diversity Improves Inter-Individual Emotion Commonality of Spatio-Spectral EEG Signatures Using Independent Component Analysis

Electroencephalogram (EEG) variability poses a great challenge to the affective brain-computer interface (aBCI) for practical applications. Most aBCI frameworks have been demonstrated successfully but deliberated on single-day data, which can be realistically susceptible to psychophysiological changes and further hinder the exploration of inter-individual EEG commonality. This study proposes a multiple-day scenario that learns exclusively from diverse EEG correlates of emotional responses on different days (i.e., enriched data diversity) by using a unified independent components analysis framework. Given an eight-day dataset of 10 subjects (i.e., 80 sessions), the results demonstrated that the multiple-day scenario intensified the inter-subject emotion commonality (i.e., the percentage of subjects with the same signature) to a certain extent when considering sufficient cross-day sessions, whereas the most commonly adopted single-day analysis (i.e., diversity-confined) led to session-dependent inferior outcomes. Given the best case, the emotional valence dimension was associated with relatively reproducible frontal beta, central midline gamma, and occipital beta modulations with 30%–40% subject commonality, whereas the arousal counterpart suffered more substantially from EEG variability and barely returned representative signatures. These results suggest that EEG signature representation may be substantially compromised by limited data diversity, impeding the efficacy and generalizability of the aBCI model in real-life settings.