Design of User-Customized Negative Emotion Classifier Based on Feature Selection Using Physiological Signal Sensors

Area of Science:

• Psychophysiology
• Affective Computing
• Biomedical Signal Processing

Background:

• Subjective emotion assessment methods (Likert scale, Self-Assessment Manikin) have limitations in capturing objective cognitive status.
• Physiological signals offer a more objective measure of emotional responses.
• Individual variations in emotional experience necessitate personalized analysis methods.

Purpose of the Study:

• To investigate the utility of physiological signals (electrocardiogram, skin temperature, electrodermal activity) for objective emotion recognition.
• To develop and validate a feature selection algorithm for enhancing emotion recognition accuracy.
• To identify key physiological markers for distinguishing negative emotions.

Main Methods:

• Utilized electrocardiogram, skin temperature, and electrodermal activity (EDA) as physiological signal inputs.
• Employed Kullback-Leibler Divergence (KLD) to analyze patterns in physiological signal distributions.
• Developed a novel feature selection algorithm to identify important physiological features for emotion classification.

Main Results:

• Electrodermal activity (EDA) features were found to be crucial for accurately distinguishing negative emotions across all participants.
• The proposed feature selection algorithm achieved an average accuracy of 92.5% in emotion recognition.
• Objective physiological measures demonstrated superior performance compared to traditional subjective methods.

Conclusions:

• Physiological signals, particularly EDA, provide a robust and objective method for emotion recognition.
• The developed feature selection algorithm significantly enhances the accuracy of negative emotion detection.
• This approach offers a promising avenue for more reliable and personalized emotion analysis.