Advanced Machine Learning Optimization of a Graphene-Based Terahertz Biosensor for Urine Glucose Detection

This article presents a tunable graphene-based terahertz (THz) biosensor for high-accuracy and label-free urine glucose detection. By exploiting the electrostatic tunability of graphene, the proposed sensor exhibits well-defined and controllable resonance shifts in response to refractive index variations associated with different urine glucose levels. The device operates within a biologically relevant refractive index range of 1.335–1.347 RIU and achieves a high sensitivity of up to 800 GHz/RIU, enabling reliable discrimination among normal, low, moderate, and high urine glucose concentrations. In addition, the biosensor demonstrates a high quality factor of 49.17 and a figure of merit (FOM) reaching 2.376 RIU−1, confirming its strong spectral selectivity and sensing efficiency. To further enhance prediction accuracy and optimize sensing performance, a machine-learning-based approach using a K-nearest neighbor (KNN) regression model is employed, yielding a high coefficient of determination (R${}^{2} = 0.988$) with reduced computational complexity. The strong sensing performance is supported by effective plasmonic field confinement inherent to the graphene-based THz biosensor. Owing to its compact design, high sensitivity, and compatibility with data-driven analysis, the proposed THz biosensor represents a promising platform for noninvasive urine glucose monitoring and future integration into intelligent biomedical sensing systems.