EEG-Based Tension Recognition Annotated with Electrodermal Activity.

Area of Science:

• Neuroscience
• Psychophysiology
• Machine Learning

Background:

• Emotion EEG dataset construction relies heavily on video stimuli.
• Current annotation methods often assign a single emotion label to entire videos, ignoring arousal fluctuations.
• This overlooks crucial variations in emotional intensity during stimulus exposure.

Purpose of the Study:

• To propose a novel method for precise emotion annotation in EEG data.
• To integrate electrodermal activity (EDA) as a psychophysiological arousal marker with EEG.
• To develop a new dataset annotated for high and low arousal states using EDA.

Main Methods:

• Collected simultaneous EEG and EDA data during video-induced emotional states.
• Utilized EDA to annotate EEG segments with high (tension) and low (calmness) arousal levels.
• Employed machine learning and deep learning algorithms in subject-specific models.

Main Results:

• Models trained exclusively on high-arousal EEG data achieved performance comparable or superior to models trained on mixed-arousal data.
• The inclusion of EDA-based arousal annotations significantly enhanced the robustness of emotion recognition.
• Subject-specific models demonstrated the effectiveness of the proposed annotation strategy.

Conclusions:

• Electrodermal activity (EDA) is a valuable tool for refining EEG-based emotion annotation.
• High-arousal EEG data alone can be sufficient for training effective emotion recognition models.
• This approach offers a more nuanced and accurate method for building emotion EEG datasets.