Automated Emotion Recognition System Using Blood Volume Pulse and XGBoost Learning

Area of Science:

• Physiological computing
• Affective computing
• Biomedical signal processing

Background:

• Emotion detection is crucial for mental health monitoring and human-computer interaction.
• Blood Volume Pulse (BVP) signals offer a non-invasive physiological measure reflecting autonomic nervous system activity.
• Existing emotion recognition methods often rely on complex multimodal data or lack real-time applicability.

Purpose of the Study:

• To develop and evaluate a machine learning model for emotion detection using Blood Volume Pulse (BVP) signals.
• To identify key features from BVP signals that are most indicative of different emotional states.
• To explore the feasibility of using wearable sensor data for emotion recognition in healthcare.

Main Methods:

• Utilized the publicly available CASE dataset comprising BVP signals from 30 subjects.
• Pre-processed BVP signals and extracted 39 features across time, frequency, and time-frequency domains.
• Developed an emotion detection model using the XGBoost algorithm, optimizing with the top 10 most significant features.

Main Results:

• The XGBoost model achieved a maximum classification accuracy of 71.88% for emotion detection.
• The most impactful features were derived from time (5), time-frequency (4), and frequency (1) domains.
• Skewness from the time-frequency representation of BVP was identified as the single most important feature for classification.

Conclusions:

• Blood Volume Pulse (BVP) signals, particularly features like skewness, hold significant potential for accurate emotion detection.
• Machine learning models, such as XGBoost, can effectively leverage BVP features for emotion classification.
• BVP signals recorded via wearable devices present a promising avenue for developing practical emotion detection systems in healthcare settings.