Abstract
Dehydration and overhydration, both have mild to severe medical implications on human health. Tracking Hydration Level (HL) is, therefore, very important particularly in patients, kids, elderly, and athletes. The limited solutions available for the estimation of HL are commonly inefficient, invasive, or require clinical trials. Need for a non-invasive auto-detection solution is imminent to track HL on a regular basis. To the best of authors' knowledge, it is for the first time a Machine Learning (ML) based auto-estimation solution is proposed that uses Galvanic Skin Response (GSR) as a proxy of HL in the human body. Various body postures, such as sitting and standing, and distinct hydration states, hydrated vs dehydrated, are considered during the data collection and analysis phases. Six different ML algorithms are trained using real GSR data, and their efficacy is compared for different parameters (i.e., window size, feature combinations etc). It is reported that a simple algorithm like K-NN outperforms other algorithms with accuracy upto 87.78% for the correct estimation of the HL.