Robust stress detection from wearable ECG using subject-specific normalization and stacking ensemble learning

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Computational Health

Background:

• Increasing mental health concerns necessitate innovative monitoring solutions.
• Existing stress detection methods often lack personalization and real-time capabilities.
• Wearable electrocardiogram (ECG) signals offer a promising avenue for continuous physiological monitoring.

Purpose of the Study:

• To develop an automated stress monitoring system using wearable ECG signals.
• To address individual physiological variability in stress detection.
• To enable real-time, personalized stress monitoring for personal medical applications.

Main Methods:

• Subject-specific normalization to account for individual differences.
• Multi-domain feature extraction including heart rate variability (HRV), P-Q-R-S-T morphology, and ECG-derived respiration (EDR).
• A Stacking Ensemble classification model integrating XGBoost, Random Forest, and Support Vector Machine (SVM).

Main Results:

• The proposed framework achieved high accuracy in stress detection.
• The Stacking Ensemble model significantly outperformed individual baseline models.
• Demonstrated robust capability for real-time stress monitoring on wearable devices.

Conclusions:

• The developed system offers a robust solution for automated, real-time stress monitoring.
• Personalized normalization and multi-domain features enhance detection accuracy.
• The framework shows significant potential for personal mental health applications using wearable ECG technology.