Physiological Signal Analysis and Stress Classification from VR Simulations Using Decision Tree Methods.

Area of Science:

• Physiological monitoring
• Machine learning for health
• Virtual reality applications

Background:

• Prolonged stress negatively impacts cardiovascular and immune systems.
• Assessing stress impact requires monitoring sympathetic (SNS) and parasympathetic (PNS) nervous systems.
• Physiological data acquisition via Electrocardiogram (ECG), Galvanic Skin Response (GSR), and Respiration (RESP) sensors is crucial.

Purpose of the Study:

• To analyze physiological stress responses during a Virtual Reality (VR) Bubble Bloom game.
• To develop and evaluate machine learning models for accurate stress classification.
• To investigate the effectiveness of novel features for stress detection.

Main Methods:

• Collected physiological data (ECG, GSR, RESP) from 15 subjects during various tasks, including VR gaming.
• Extracted 21 features and developed a novel K-means feature from 11 others.
• Utilized personalized Classification and Regression Tree (CART), Decision Tree (DT), Ensemble Gradient Boosting (EGB), and Extreme Gradient Boosting (XGBoost) models.

Main Results:

• Heart rate (HR), GSR, and RESP decreased, while high frequency (HF) increased post-VR game.
• Personalized CART achieved 87.75% accuracy for binary stress classification.
• EGB model reached 100% accuracy for binary stress classification, outperforming existing studies.
• XGBoost and DT models classified five stress classes with 72.22% accuracy using the novel K-means feature.

Conclusions:

• The proposed methods, particularly EGB, demonstrate high efficacy in classifying stress levels.
• The novel K-means feature improves model performance and reduces errors in stress classification.
• VR gaming combined with advanced machine learning offers a promising approach for stress management and assessment.