IoT and cloud-based non-invasive diabetes detection system from photoplethysmogram

Diabetes is a widespread chronic metabolic disorder that affects a substantial portion of the global population and requires diligent management and continuous medical supervision. Traditional glucose monitoring methods often involve invasive procedures that cause discomfort and increase the risk of infection. With the rapid advancements of the Internet of Things (IoT) and cloud computing, healthcare is evolving for early disease detection and personalized treatment solutions. The aim of this research is to develop a system based on the Internet of Things (IoT) for non-invasive diabetes detection using Photoplethysmography (PPG) signals. Our system leverages machine learning models, such as XGBoost, Random Forest, Logistic Regression, and SVM, to accurately classify diabetic patients using statistical features collected from PPG signals and metadata. To ensure model robustness, we employed a fivefold cross-validation approach. The hardware technology includes a portable, cost-effective system with a simple NIR sensor and a high-performing ESP32 module that minimizes user discomfort and ensures easy integration. Deployed on the Amazon Web Services (AWS) cloud platform, our system enables real-time data collection and analysis, providing users with a convenient web application for monitoring their diabetic status. The performance of the developed system has been tested with two different datasets: the PPG-BP dataset and the Mazandaran dataset, achieving 87.88% and 90% accuracy, respectively. The results demonstrate the potential of the designed system to revolutionize diabetes management by providing a non-invasive, accessible, and accurate solution for early-stage detection.