Emotion recognition from multimodal physiological measurements based on an interpretable feature selection method.

Area of Science:

• Psychophysiology
• Machine Learning
• Biomedical Signal Processing

Background:

• Physiological signals are used for emotion classification, but lack interpretability.
• Existing methods struggle with feature-specific characterizations for emotional states.

Purpose of the Study:

• To propose a feature selection method for physiological signals to enhance emotion classification interpretability.
• To identify key physiological features discriminating arousal and valence dimensions of emotions.

Main Methods:

• Developed a feature selection technique preserving the original dimensional space of physiological signals.
• Utilized Galvanic Skin Response (GSR) and Electrocardiogram (ECG) signals.
• Employed Linear Discriminant Analysis (LDA) with top-ranked features.

Main Results:

• GSR features are crucial for arousal discrimination (fear vs. happiness/relaxation).
• GSR signal's average/median and ECG's SD1/SD2 ratio are key for valence discrimination.
• LDA model achieved high accuracies: 72% (happiness), 67% (relaxation), 89% (fear).

Conclusions:

• Physiological signals can effectively assess emotional states.
• The proposed method offers a fast, efficient way to select important autonomic nervous system indexes.
• Potential for clinical extension to stress, pain, PTSD, and depression assessment.