Global Stress Detection Framework Combining a Reduced Set of HRV Features and Random Forest Model

Area of Science:

• Biomedical Engineering
• Data Science
• Psychophysiology

Background:

• Chronic stress significantly impacts global adult health, contributing to serious conditions like heart disease and anxiety.
• Existing stress detection methods using wearable devices lack standardized feature sets and often rely on person-specific training.
• Wearable wristbands are widely adopted, presenting an opportunity for improved stress monitoring.

Purpose of the Study:

• To develop and validate a global stress detection model using a unified set of features and a machine learning algorithm.
• To investigate the efficacy of combining eight heart rate variability (HRV) features with a random forest (RF) algorithm for stress detection.
• To establish a stress detection model that can be trained globally but evaluated for individual performance.

Main Methods:

• Selected eight optimal HRV features using the minimum redundancy maximum relevance (mRMR) method for efficient model training.
• Implemented a global training framework for the random forest (RF) algorithm, incorporating data from all subjects.
• Validated the proposed global stress model using the WESAD and SWELL open-access databases, individually and combined.

Main Results:

• The global stress monitoring model achieved an accuracy exceeding 99% in identifying person-specific stress events.
• The mRMR method effectively reduced training time by selecting the most informative HRV features.
• The model demonstrated robust performance across different datasets, confirming its generalizability.

Conclusions:

• A global stress detection model trained on diverse data can accurately identify individual stress events.
• The integration of selected HRV features and RF algorithm offers a promising approach for non-invasive, high-accuracy stress monitoring.
• Future research should focus on real-world application testing of this global stress monitoring framework.