Abstract
There is a critical need to design noninvasive, user-friendly hydration assessment tools across diverse population groups. In this article, we capitalize on the capacitive nature of smartphone screens for in situ dehydration monitoring in two population groups, i.e., fasting people and sportspeople, using smartphones. We utilize the FDC2114 board with its two capacitive sensors to emulate a smartphone touchscreen and demonstrate that the capacitive sensing could reliably differentiate between five distinct hydration levels of a fasting person and do dehydration detection for sportspeople. Our methodology involves collecting data from 35 fasting subjects five times a day between morning and evening at regular intervals, and from ten sportspeople before and after sports/exercise activity. This is followed by preprocessing, and 5-class, 4-class, and binary classification using a number of machine learning (ML) models. The results show that the linear logistic regression (LLR) model outperforms all other models for each population group, which demonstrates the sensitivity of capacitive sensing to the body's dielectric properties that vary with hydration levels. The study also reveals distinct dehydration patterns during fasting and exercising, which is mainly due to the sweat phenomenon that is more prominent in the sportspeople group. Therefore, we are of the opinion to train and fine-tune ML models for each population group separately.
Original language | English (US) |
---|---|
Pages (from-to) | 11428-11440 |
Number of pages | 13 |
Journal | IEEE Sensors Journal |
Volume | 25 |
Issue number | 7 |
DOIs | |
State | Published - 2025 |
Bibliographical note
Publisher Copyright:© 2025 IEEE.
Keywords
- Dehydration
- fasting
- machine learning (ML)
- skin capacitance
- sportspeople
ASJC Scopus subject areas
- Instrumentation
- Electrical and Electronic Engineering