Classification of Perceived Human Stress using Physiological Signals.

Area of Science:

• Physiological computing
• Biomedical signal processing
• Machine learning for health

Background:

• Accurate stress detection is crucial for mental well-being and performance.
• Non-invasive physiological signals offer a promising avenue for objective stress assessment.
• Existing methods often rely on stress inducers, limiting real-world applicability.

Purpose of the Study:

• To investigate the efficacy of non-invasive physiological signals for classifying perceived human stress.
• To develop and evaluate a classification scheme using electroencephalography (EEG), galvanic skin response (GSR), and photoplethysmography (PPG).
• To compare the performance of different machine learning classifiers for stress detection.

Main Methods:

• Acquired physiological data (EEG, GSR, PPG) from 28 participants under open-eye conditions.
• Extracted four time-domain features from each signal type.
• Employed Support Vector Machine (SVM), Naive Bayes, and Multi-Layer Perceptron (MLP) classifiers for stress classification.

Main Results:

• The Multi-Layer Perceptron (MLP) classifier achieved the highest classification accuracy of 75%.
• The proposed classification scheme demonstrated superior performance compared to existing methods.
• Effective stress classification was achieved without the use of artificial stress inducers.

Conclusions:

• Non-invasive physiological signals, particularly when analyzed with advanced machine learning models like MLP, are effective for perceived stress classification.
• The developed method offers a practical and non-intrusive approach to stress monitoring.
• This research provides a foundation for developing real-time stress management systems.