EEG-Based Tension Recognition Annotated with Electrodermal Activity

Wei Gu, Yun-Huan Li, Li-Ming Zhao, Wei-Long Zheng, Bao-Liang Lu

Abstract: The precise annotation of emotions is crucial for constructing emotion EEG datasets, where videos are the dominant emotion-inducing tools. However, existing annotation methodologies are often limited, assigning a uniform label to the entire video and neglecting subjects' emotional arousal variations during viewing. This paper proposes a novel approach to address this issue by integrating electrodermal activity (EDA), a psychophysiological marker of arousal, with EEG data. We introduce a new dataset that captures both tension and calmness, utilizing EDA to annotate EEG data with high and low arousal. The method is systematically tested in subject-specific paradigms, employing a suite of machine learning and deep learning algorithms. Our results demonstrate that models trained solely on highly induced EEG data, comprising 71.75% of the initial training set, yield equal or superior performance on test sets, regardless of their arousal levels. This underscores the potential of EDA in enhancing emotion recognition accuracy in EEG studies.