Non-invasive Blood Glucose Estimation Using Statistical Features Defined via Convex Combination of One Norm and Infinity Norm Optimization Problems

Abstract

In the past few decades, due to the increasing emphasis on health, blood glucose, a healthy reference value, has received more and more attention. Traditional invasive blood glucose testing methods require pricking a finger to take a drop of blood, and measuring blood glucose levels based on how the device reacts with the blood. Due to various shortcomings of traditional methods, and the semi-invasive or minimally invasive blood glucose monitoring systems that have been marketed in many countries and regions have high costs and some usage limitations, a new type of easy-to-use non-invasive blood glucose detection and prediction system is rapidly developing. This paper introduces a wearable non-invasive blood glucose detection device using near-infrared technology and its data processing technology, which includes extracting features from the obtained signals and using machine learning methods for blood glucose level prediction, and novel use of the solution The optimization problem of different norm values is used to obtain new statistical features to further improve the accuracy of non-invasive blood glucose prediction.