HydrateMe: A data-driven hydration monitoring model and application using machine learning approach

Accurate hydration monitoring is critical for health and performance but remains limited by sparse, reliance on isolated weight-based or sweat-based signal data and unlabelled datasets. This study re-conditioned the publicly available SpectroPhon DBM dataset, using physiologically grounded algorithms, to extrapolate hydration rate and sweat osmolality ultimately leading to the generation of sweat-based binary labels as well as binary and multiclass weight-based labels, thus enriching the dataset for machine learning tasks. Among the tested models (Random Forest, Support Vector Machine, XGBoost, and Logistic Regression), XGBoost achieved 100% accuracy in sweat-based binary classification and 97.56% in weight-based four-class tasks; Random Forest reached 100% in weight-based binary classification. As a practical demonstration, a Random Forest model trained on binary sweat-based label was integrated into a Gradio-based user interface for monitoring hydration. The resulting application, named HydrateMe, offers a proof of concept for real-time, user-friendly hydration monitoring, aimed at supporting clinicians, athletes, and caregivers. Future iterations will incorporate real-time sensor data and Bluetooth streaming for personalised, continuous assessment. The research shows the value of combining public data, domain-driven feature enrichment, and ML to advance non-invasive hydration monitoring. It provides a scalable foundation for practical applications in healthcare, sports science, and occupational safety.