Exploring day-to-day variability in EEG-based emotion classification

The research of electroencephalography (EEG)-based emotion classification has gained much popularity in the past few years. Researchers continue to seek an optimal machine learning-based pipeline to characterize the associations between complex spatio-spectral EEG dynamics and implicit emotional responses. However, toward a real-life application, addressing the inherent day-to-day variability in EEG signals is also of urgent importance, yet was less concerned in the literature. This study explored the day-to-day EEG variability and tested the feasibility of developing an emotion-classification pipeline that can account for such variability. The empirical results of this study showed that the use of a proper feature extraction, e.g., band-power asymmetries over the fronto-posterior regions, in conjunction with an effective artifact removal method, e.g., independent component analysis, could alleviate the impacts of inter-day variability and improve the classification performance.