Non-Invasive Blood Glucose Estimation System Using 660 nm Red and 880 nm IR Wavelengths

—This project describes the development of a multichannel photoplethysmography (PPG) system for non-invasive blood glucose detection. Addressing the rise of diabetes mellitus and the discomfort of traditional invasive testing, this research offers a pain-free alternative. The device utilizes a dual-channel PPG sensor with red and infrared (IR) LEDs to capture physiological signals. Key features from these signals are then extracted and processed using machine learning to estimate blood glucose levels. A decision tree-based regression model, specifically using the XGBoost (XGB) algorithm, is constructed to predict glucose concentrations by correlating PPG signal attributes with data from conventional glucose meters. The system’s accuracy is rigorously evaluated using standard metrics like Mean Absolute Error (MAE), Root-Mean-Squared Error (RMSE), Pearson’s r, and Clarke Error Grid Analysis (CEGA). The research culminates in a compact, portable device for convenient and regular self-monitoring.