Automated Stress Recognition Using Supervised Learning Classifiers by Interactive Virtual Reality Scenes.

Area of Science:

• Psychology
• Biomedical Engineering
• Computer Science

Background:

• Virtual reality (VR) offers novel therapeutic avenues for stress disorders.
• Objective physiological measures can potentially validate subjective stress experiences.

Purpose of the Study:

• To develop and evaluate a VR stress training system for stress recognition.
• To correlate physiological data with self-reported stress levels.
• To automate stress recognition using machine learning models.

Main Methods:

• A VR system with three interactive stressful scenes was developed.
• Physiological data (heart rate, electrodermal activity, eye-blink rate) and self-reported stress (SSAI questionnaire) were collected.
• Supervised learning models, including bagging ensemble learning and multivariate stepwise regression, were applied for stress classification and prediction.

Main Results:

• The VR system demonstrated reliability (Cronbach's alpha = 0.72, KMO = 0.78) and validity through factor analysis.
• Physiology-based stress level classification achieved 0.742 accuracy using a bagging ensemble model.
• Multivariate stepwise regression showed a goodness-of-fit of 0.44 for continuous stress score prediction.

Conclusions:

• The VR stress training system effectively elicits stress and correlates physiological responses with subjective ratings.
• Automated stress recognition using physiological data is feasible and accurate.
• This technology supports further research in stress disorder recognition and treatment validation.