Assessment of Valance Emotional State Using EEG-EDA Coupling and Explainable Classifiers

Area of Science:

• Neuroscience
• Affective Computing
• Biomedical Signal Processing

Background:

• Emotions significantly influence daily life, involving complex interactions between central and peripheral nervous system signals.
• Existing emotion classification methods often focus on unimodal or simple multimodal approaches, neglecting the intricate interplay of physiological signals.
• There is a need for advanced methods that integrate central nervous system (Electroencephalogram - EEG) and peripheral nervous system (Electrodermal Activity - EDA) signal interactions for accurate emotion characterization.

Purpose of the Study:

• To investigate the utility of coupling Electroencephalogram (EEG) and Electrodermal Activity (EDA) signals for characterizing valence emotional states.
• To develop and evaluate a multimodal approach that combines EEG-derived features, EDA-derived features, and their inter-signal coupling features for emotion classification.
• To identify the most effective features and classifiers for distinguishing between different valence emotional states using physiological signal interactions.

Main Methods:

• Utilized the publicly available DEAP database comprising EEG and EDA signals from 32 subjects.
• Decomposed EEG signals into frequency bands (θ, α, β, γ) and EDA signals into phasic and tonic components.
• Extracted EEG, EDA, and EEG-EDA coupling features, applying Random Forest (RF), Linear Discriminant Analysis, and Adaptive Boosting classifiers, complemented by SHAP analysis for feature interpretability.

Main Results:

• The combined feature set of EEG, EDA, and their coupling, processed by a Random Forest classifier, achieved the highest F1-score of 68.21% for valence emotion classification.
• SHAP analysis indicated that features derived from frontal electrodes, particularly in the gamma (γ) band, provided significant discrimination among different valence states.
• The proposed approach demonstrated the capability to classify valence emotional states effectively by leveraging the interaction between EEG and EDA signals.

Conclusions:

• The coupling of EEG and EDA signals offers a promising avenue for enhancing the accuracy of emotion classification.
• The integration of central and peripheral nervous system signal interactions provides a more comprehensive understanding of emotional processes.
• This methodology holds potential for application in clinical settings for objective emotional state assessment.