Human Stress Classification Using Cardiovascular and Respiratory Data Based on Machine Learning Techniques

Area of Science:

• Physiological computing
• Biomedical signal processing
• Machine learning in healthcare

Background:

• Stress significantly impacts mental health, well-being, and performance.
• Current stress detection methods often rely on single physiological signals.
• A novel approach integrating cardiovascular and respiratory data for stress detection is proposed.

Purpose of the Study:

• To develop and validate a multimodal system for accurate, real-time stress detection.
• To investigate the efficacy of fusing cardiovascular and respiratory signals for stress classification.
• To identify key physiological features that discriminate stress states.

Main Methods:

• Collected electrocardiogram (ECG) and respiration signals from 15 participants during a stress induction task.
• Developed a real-time peak detection algorithm for ECG signal processing.
• Extracted time and frequency domain features from both signal types and employed machine learning models (SVM, KNN, bagged trees, random forests) for classification, validated with NASA-TLX scores.

Main Results:

• Multimodal data fusion significantly improved stress classification accuracy to 95.6% ±1.7% compared to single modalities.
• Forward feature selection identified crucial discriminative features from both cardiovascular and respiratory signals.
• The developed system demonstrated superior performance over single-modality approaches.

Conclusions:

• Integrating multimodal physiological data, specifically cardiovascular and respiratory signals, is highly effective for accurate stress detection.
• This approach surpasses the performance of single-modality methods and existing literature.
• Findings support the development of real-time monitoring systems for enhanced stress and health management.