Exploring Day-to-Day Variability in the Relations Between Emotion and EEG Signals

Electroencephalography (EEG)-based emotion classification has drawn increasing attention over the last few years and become an emerging direction in brain-computer interfaces (BCI), namely affective BCI (ABCI). Many prior studies devoted to improve emotion-classification models using the data collected within a single session or day. Less attention has been directed to the day-to-day EEG variability associated with emotional responses. This study recorded EEG signals of 12 subjects, each underwent the music-listening experiment on five different days, to assess the day-to-day variability from the perspectives of inter-day data distributions and cross-day emotion classification. The empirical results of this study demonstrated that the clusters of the same emotion across days tended to scatter wider than the clusters of different emotions within a day. Such inter-day variability poses a severe challenge for building an accurate cross-day emotion-classification model in real-life ABCI applications.