A Hexagonal AMC Backed – Circular Microstrip Patch Antenna for Noninvasive Blood Glucose Monitoring: An In‐Silico and In‐Vitro Based Experimental Study Using Artificial Intelligence

RF and Microwaves‐based sensors are gaining researchers' attention because of their non‐invasive and non‐ionizing nature for diabetes management, and affordable nature of this technology with no side effects. This article, therefore, proposes a flexible circular patch antenna with diamond wand shape, as a radiator for sensing changes in BGL in correspondence to S parameter variation. To improve the gain and front‐to‐back ratio of the proposed radiator, it is backed with a hexagonal‐shaped metasurface. The proposed sensor (Antenna + Metasurface) is then simulated with a three‐layered bio‐phantom with varying BGL, and the S ‐parameter data is captured in 3D EM software, CST MWS EM simulator. The proposed sensor resonates at 2.45 GHz, covering the Industry Scientific Medical band of 2.4–2.5 GHz with a gain of 8.09 dBi and a specific absorption rate of 0.0008 W/Kg. The validation of the proposed sensor is done by fabricating (using a wet etching process) and testing the proposed antenna and AMC for their radiation properties using a VNA (Keysight E 5063) and an anechoic chamber. The BGL detection procedure is validated by preparing aqueous glucose solutions of different glucose concentrations (100, 200, 300 and 400 mg/dL) and comparing the measured and simulated S parameter values for these cases. Additionally, to accurately predict the blood glucose values using S 11 , regression analysis of the S parameter data is carried out using Machine Learning algorithms such as Support Vector Regression, Random Forest, Extreme Gradient Boosting, and Optuna‐tuned Cat Boost (OtCB) models. The OtCB model showed the best performance with a R 2 score of 0.9982 and a mean absolute error of 0.0742, and hence is best suited for predicting BGL values for the current usage. The models demonstrated a median absolute error below 0.3 dB even after synthetic noise injection, which shows their robustness for real‐life practical scenarios.