Machine Learning-Integrated Microwave Sensing of Glucose Concentration: A Classification Approach

In this work, an integrated approach based on the combination of microwave sensors and machine learning techniques is presented, to be exploited in the framework of continuous and non-invasive monitoring of blood glucose concentration. In particular, by employing a non-parametric classification approach, some significant features (i. e. the resonance frequency, the maximum value, the −10 dB bandwidth and the equivalent signal energy) included in the frequency response of microwave resonant sensors are used as predictor variables, while a discretized version of the glucose concentration is used as response variable. The discussed numerical results confirm the validity of the proposed approach, which allows to obtain a satisfactory interval estimation of the glucose concentration.