Blood Glucose Level Prediction Using ARMAX and ARMAX-ANN Models

Abstract

Diabetes is a serious medical condition that can lead to complications such as stroke, heart disease, blindness, and obesity. An estimated 347 million people were affected by diabetes, with approximately 3.4 million deaths attributed to high blood sugar levels. Researchers have explored various non-invasive techniques to measure blood glucose levels, including ultrasonic sensors, multisensory systems, absorbance and transmittance methods, bio-impedance, voltage intensity, and thermography. The development of non-invasive glucose monitoring methods continues to be a significant area of interest in the medical field. This study investigates the application of near-infrared (NIR) spectroscopy for glucose level measurement, alongside the use of linear and non-linear system identification models to predict output data from NIR measurements. While NIR spectroscopy has been utilized in previous studies, the optimal wavelength range remains a topic of debate among researchers. To assess the feasibility of a linear approach, the Autoregressive Moving Average Exogenous (ARMAX) model was applied to predict NIR measurement outcomes. Additionally, a non-linear model combining ARMAX with an Artificial Neural Network (ANN) was implemented, allowing for a comparative analysis of the performance of both linear and non-linear prediction methods