Optimal Electrodermal Activity Segment for Enhanced Emotion Recognition Using Spectrogram-Based Feature Extraction...

Area of Science:

• Physiological computing
• Affective computing
• Biomedical signal processing

Background:

• Accurate emotion recognition from physiological signals is crucial for clinical and scientific applications.
• Electrodermal activity (EDA) is a key physiological indicator of emotional arousal.
• Previous research highlights the importance of signal segmentation for robust emotion recognition systems.

Purpose of the Study:

• To optimize the segmentation of electrodermal activity (EDA) signals for improved emotion recognition.
• To evaluate the performance of different EDA signal segments in classifying emotions.
• To determine the most effective EDA segment for emotion recognition systems.

Main Methods:

• Acquired EDA signals from the CASE and WESAD datasets for dimensional and categorical emotion classification.
• Pre-processed and decomposed EDA signals into phasic components using convex optimization.
• Segmented phasic signals, generated spectrograms, and extracted 85 features.
• Developed and compared four machine learning models using whole, first-part, and second-part phasic EDA signals.

Main Results:

• The CASE dataset yielded a maximum multi-class accuracy of 62.54% with whole phasic signals and 61.75% with the second part.
• The WESAD dataset achieved 96.44% accuracy for three-class emotion classification using both whole and second-part phasic segments.
• The second part of the EDA signal demonstrated strong performance across both datasets and emotion classification types.

Conclusions:

• The second part of the electrodermal activity (EDA) signal is recommended for optimal emotion recognition.
• Effective EDA signal segmentation significantly enhances the performance of emotion recognition systems.
• This study provides valuable insights for developing more accurate and reliable emotion recognition technologies.