Deep Learning Framework for Categorical Emotional States Assessment Using Electrodermal Activity Signals

Area of Science:

• Affective computing
• Physiological computing
• Machine learning for emotion recognition

Background:

• Accurate classification of emotional states is crucial for understanding human behavior and mental health.
• Electrodermal Activity (EDA) signals offer a promising physiological measure for emotion detection.
• Developing robust machine learning models for real-time emotion classification remains a challenge.

Purpose of the Study:

• To classify categorical emotional states (amusing, boring, relaxing, scary) using Electrodermal Activity (EDA) signals.
• To develop and evaluate a configurable Convolutional Neural Network (cCNN) model for emotion discrimination.
• To assess the pipeline's effectiveness in analyzing emotional states in both normal and clinical contexts.

Main Methods:

• Utilized the publicly available Continuously Annotated Signals of Emotion dataset.
• Processed EDA signals: down-sampling, decomposition into phasic components via cvxEDA, and Short-Time Fourier Transform for spectrogram generation.
• Employed a configurable Convolutional Neural Network (cCNN) with nested k-Fold cross-validation for model training and evaluation.

Main Results:

• The proposed pipeline achieved high average classification performance.
• Key metrics included: 80.20% accuracy, 60.41% recall, 86.8% specificity, 60.05% precision, and 58.61% F-measure.
• Demonstrated the model's robustness and ability to discriminate between distinct emotional states.

Conclusions:

• The developed pipeline effectively classifies categorical emotional states using EDA signals and a cCNN.
• The findings suggest the pipeline's potential utility in the examination of diverse emotional states.
• This approach could be valuable for applications in both typical and clinical psychological assessments.