Explainable stress type classification captures physiologically relevant responses in the Maastricht Acute Stress...

Area of Science:

• Physiology
• Machine Learning
• Artificial Intelligence

Background:

• Current stress detection methods often fail to distinguish between different types of stress.
• There is a need for more specific and explainable stress classification systems.

Purpose of the Study:

• To develop and evaluate a multimodal machine learning approach for classifying distinct physiological stress responses.
• To investigate the impact of signal measurement and feature selection on stress classification accuracy.
• To apply explainable AI (XAI) to understand the physiological basis of stress detection.

Main Methods:

• Physiological responses were measured during the Maastricht Acute Stress Test (MAST), involving cold pressor and mental arithmetic tasks.
• Mixed-effects models were used to compare responses to baseline and between stressor types.
• Machine learning models were trained and evaluated, with XAI used to interpret feature importance.

Main Results:

• The model successfully distinguished between different stressor types and baseline with up to 86.5% balanced accuracy.
• The selection of physiological signals and features significantly impacted classification performance.
• XAI analysis confirmed that the model focused on physiologically relevant features for each stressor.

Conclusions:

• Multimodal machine learning can accurately differentiate various stress reactions based on physiological changes.
• Signal and feature selection are critical determinants of stress detection performance.
• The study provides a foundation for more nuanced and explainable stress monitoring systems.