Microwave Glucose Concentration Classification by Machine Learning

This work aims to utilize machine learning algorithms to classify glucose concentration from the measured broadband microwave scattering signals (S-11). The sweeping frequency signals are first measured from glucose aqueous solution with various concentrations from pure water to 1000 mg/dL. Dielectric parameters are then extracted based on the modified Debye dielectric dispersion model and utilized as the features to create a larger dataset by adding Gaussian noises at various levels. Two separate datasets are created; one containing S-11 parameters and another containing Debye dielectric parameters. Several machine learning algorithms are used to classify glucose concentrations. Results indicate that the best algorithm can achieve perfect glucose concentration classification accuracy for the Debye dielectric parameter-based feature sets. The study suggests an alternative way to develop the non-invasive glucose detection method using machine learning.