Non-invasive Aqueous Glucose Monitoring using Microwave Sensor with Machine Learning

Abstract

In this paper, a microwave dipole sensor is used with machine learning algorithms to build a non-invasive CGM system. Preliminary, the sensor is used on aqueous (water-glucose) solutions with different glucose concentrations to check the sensitivity of the sensor to those different glucose concentrations. Machine learning is used to extract the appropriate features from the signals coming from the aqueous solutions and then to build the regression model using those extracted features to be able to predict the actual glucose levels. Using more than 19 regression models, the results showed a good accuracy with Root Mean Square Error $\mathbf{(RMSE)=6.7}$ by Matern 5/2 Gaussian Process Regression (GPR) algorithm. More data is needed to improve the accuracy of the prediction.