Liu Y.-H.CHIEN-TE WUKao Y.-H.Chen Y.-T.2020-04-082020-04-08201397814577021671557-170Xhttps://www.scopus.com/inward/record.uri?eid=2-s2.0-84886467899&doi=10.1109%2fEMBC.2013.6610498&partnerID=40&md5=b8791e8f52ce633fc68ea024d262c1c6https://scholars.lib.ntu.edu.tw/handle/123456789/483702Single-trial electroencephalography (EEG)-based emotion recognition enables us to perform fast and direct assessments of human emotional states. However, previous works suggest that a great improvement on the classification accuracy of valence and arousal levels is still needed. To address this, we propose a novel emotional EEG feature extraction method: kernel Eigen-emotion pattern (KEEP). An adaptive SVM is also proposed to deal with the problem of learning from imbalanced emotional EEG data sets. In this study, a set of pictures from IAPS are used for emotion induction. Results based on seven participants show that KEEP gives much better classification results than the widely-used EEG frequency band power features. Also, the adaptive SVM greatly improves classification performance of commonly-adopted SVM classifier. Combined use of KEEP and adaptive SVM can achieve high average valence and arousal classification rates of 73.42% and 73.57%. The highest classification rates for valence and arousal are 80% and 79%, respectively. The results are very promising.conference paper10.1109/EMBC.2013.6610498241106852-s2.0-84886467899