Jao, Ping KengPing KengJaoLin, Yuan PinYuan PinLinYI-HSUAN YANGJung, Tzyy PingTzyy PingJung2023-10-202023-10-202015-11-0497814244927181557170Xhttps://scholars.lib.ntu.edu.tw/handle/123456789/636333An emerging challenge for emotion classification using electroencephalography (EEG) is how to effectively alleviate day-to-day variability in raw data. This study employed the robust principal component analysis (RPCA) to address the problem with a posed hypothesis that background or emotion-irrelevant EEG perturbations lead to certain variability across days and somehow submerge emotion-related EEG dynamics. The empirical results of this study evidently validated our hypothesis and demonstrated the RPCA's feasibility through the analysis of a five-day dataset of 12 subjects. The RPCA allowed tackling the sparse emotion-relevant EEG dynamics from the accompanied background perturbations across days. Sequentially, leveraging the RPCA-purified EEG trials from more days appeared to improve the emotion-classification performance steadily, which was not found in the case using the raw EEG features. Therefore, incorporating the RPCA with existing emotion-aware machine-learning frameworks on a longitudinal dataset of each individual may shed light on the development of a robust affective brain-computer interface (ABCI) that can alleviate ecological inter-day variability.enconference paper10.1109/EMBC.2015.7318426267363262-s2.0-84953311199https://api.elsevier.com/content/abstract/scopus_id/84953311199