PSAML: A Methodological Approach for Noninvasive Computerized Hydration Level Estimation.

Area of Science:

• Physiological monitoring
• Biomedical engineering
• Machine learning applications

Background:

• Accurate hydration level (HL) monitoring is vital for health, sports, and wellness.
• Current methods for HL assessment are often invasive or noninvasive but lack accuracy and convenience.
• There is a need for efficient, noninvasive methods for real-time HL estimation.

Purpose of the Study:

• To propose and validate a novel noninvasive method for estimating hydration level (HL) using galvanic skin response (GSR) signals.
• To integrate principal component analysis (PCA), successive decomposition index (SDI), and machine learning (ML) classifiers for enhanced HL estimation.
• To demonstrate the efficacy of the proposed PSAML approach for wearable HL monitoring.

Main Methods:

• Collected GSR data from healthy subjects in various hydration states (dehydrated, normal, overhydrated) and postures.
• Preprocessed GSR signals including outlier removal, filtering, and time-window segmentation.
• Extracted and compared conventional, PCA-based, and SDI-based features using six ML classifiers.

Main Results:

• Conventional time-domain features yielded a maximum accuracy of 63.97%.
• PCA-based feature reduction significantly improved classification accuracy, with PCA+SVM, PCA+LR, and PCA+LDA achieving >99% accuracy.
• SDI-based features showed strong performance with specific classifiers in smaller time windows.

Conclusions:

• The proposed PSAML approach, integrating PCA, SDI, and ML, offers a highly accurate and efficient solution for noninvasive HL monitoring.
• This method overcomes limitations of existing HL assessment techniques, enabling reliable wearable applications.
• The findings support the potential of GSR-based analysis for advanced physiological status assessment.