NIR-Spectroscopic Classification of Blood Glucose Level using Machine Learning Approach

Diabetes Mellitus (DM) or diabetes is one of the metabolic diseases exhibiting high blood glucose level over a prolonged period. The management of diabetes is associated with the proper monitoring of blood glucose level. Researchers have been working on developing robust techniques to monitor the level of glucose in the blood. This paper aims at predicting blood glucose levels based on NIR spectroscopic response data utilizing machine learning techniques. Blood glucose samples were prepared in a controlled environment and the NIR spectrums of the samples were obtained using NeoSpectraMicro development kit. Two machine learning approaches have been employed to analyze the experimental dataset. Firstly, the Random Forest Algorithm (RF) followed by Support Vector Machine (SVM) has been utilized that provides an accuracy of 67.5%. Then, a combination of Principle Component Analysis (PCA) and SVM is used. PCA followed by SVM shows a promising result of 77.5% accuracy compared to the previous technique. The numerical findings reveal that the NIR spectroscopy with appropriate data modeling algorithm can be a potential candidate for non-invasive blood glucose monitoring system.