Enhanced PPG-based stress recognition: a transfer learning approach to internal vs. external stress

Area of Science:

• Physiology
• Computer Science
• Machine Learning

Background:

• Stress recognition is crucial for mental and physical health monitoring.
• Existing methods struggle to differentiate between various types of stress.
• Photoplethysmographic (PPG) signals offer a non-invasive method for physiological monitoring.

Purpose of the Study:

• To create a comprehensive dataset for internal and external stress assessment.
• To develop automated stress recognition methods using PPG signals.
• To enhance the accuracy and robustness of stress detection models.

Main Methods:

• Established the Internal and External Stress Dataset (IESD) with PPG signals from 107 participants.
• Developed a transfer learning model using Domain Adversarial Neural Network (DANN) and Maximum Mean Discrepancy (MMD).
• Employed deep feature extraction for improved recognition performance.

Main Results:

• Identified significant differences in heart rate variability (HRV) between internal and external stress paradigms.
• Achieved high accuracy in recognizing homologous stress (e.g., 73.86% from TSST to ST).
• Demonstrated improved recognition performance and robustness across different stress contexts.

Conclusions:

• The IESD and proposed deep learning model enhance automated stress detection capabilities.
• The methodology effectively differentiates internal and external stress.
• Deep learning significantly improves stress recognition accuracy for future applications.