Development of non-invasive glucose monitoring system : integration of spectroscopy, IoT, and machine learning for real-time health insight

Abstract

The prevalence of diabetes worldwide has necessitated the development of affordable, accessible, and non-invasive glucose monitoring devices. This thesis presents the design and development of a non-invasive blood glucose monitoring system using multi-wavelength IR spectroscopy combined with machine learning techniques. A novel multi-wavelength approach was proposed, targeting specific NIR wavelengths (940 nm, 1410 nm, 1550 nm, and 1610 nm) to detect glucose changes while accounting for tissue calibration and hydration interference. The system integrates embedded hardware components, an LED driving and photodiode detection circuit, and real-time data acquisition with a web dashboard and mobile application for monitoring. Machine learning models were planned for glucose level prediction based on system-collected data. Calibration experiments using glucose solutions were conducted to validate the effectiveness of the selected wavelengths. Results demonstrate promising voltage response trends and reliable calibration curve fitting, highlighting the potential of multi-wavelength NIR sensing for non-invasive glucose monitoring. The study discusses system limitations and proposes future improvements to enhance device performance and clinical applicability.