Glucose measurement via noninvasive methods

Abstract

Noninvasive glucose measurement methods are wide-ranging; they use several different technologies to try and get accurate results. Some try to measure glucose through lacerations on the skin and chemicals, others try to do it analyzing the color of the sclera, and others try to do it analyzing the sweat. For this thesis, a completely noninvasive and chemical free approach is used. Glucose levels are classified into three useful categories (low, medium, and high) trough the use of machine learning and descriptors from chaos theory to obtain a a satisfactory Support Vector Machine (SVM) model. Several classification models are compared by the following metrics: Area Under the Receiver (AUC), accuracy, precision, recall, and their combined information (F1). And lastly, a multipurpose system that uses principles from Internet of Things (IoT) is implemented to integrate a sampling device powered by Arduino with a web app, which in turn uses cloud computing to process data and store it in a remote server to effectively train machine learning models written in Python.