Mild Dehydration Identification Using Machine Learning to Assess Autonomic Responses to Cognitive Stress

Area of Science:

• Physiology
• Biomedical Engineering
• Human Performance

Background:

• Mild dehydration poses challenges for physiological monitoring.
• Autonomic nervous system (ANS) responses are sensitive to hydration status.
• Cognitive stress tasks can reveal subtle physiological changes.

Purpose of the Study:

• To investigate the feasibility of detecting mild dehydration using autonomic responses.
• To evaluate the effectiveness of various machine learning classifiers for dehydration detection.
• To identify key autonomic indices indicative of dehydration.

Main Methods:

• Subjects (n=17) underwent cognitive stress via the Stroop task under hydrated and mildly dehydrated conditions.
• Autonomic control indices from electrodermal activity (EDA) and pulse rate variability (PRV) were measured.
• Nine classification approaches, including SVM, KNN, and ensemble methods, were tested using cross-validation.

Main Results:

• High accuracy (91.2%) was achieved in detecting mild dehydration using cubic SE-KNN with rest-stage indices.
• Similar accuracy (91.2%) was observed with cubic SVM classifiers using test-stage indices.
• Rest-to-test increments in autonomic indices yielded 86.8% accuracy with KNN and QDA classifiers.

Conclusions:

• Autonomic function measures derived from EDA and PRV are effective for detecting mild dehydration.
• Noninvasive testing of dehydration status is potentially achievable using these physiological markers.
• Cognitive stress paradigms combined with machine learning offer a promising approach for hydration monitoring.