Machine learning-enhanced graphene-gold hybrid metasurface sensor for high-precision terahertz detection of Isoquercitrin biomarkers

Abstract

This study presents a sensor operating in the terahertz (THz) frequency range for the selective detection and quantification of isoquercitrin, a crucial flavonoid biomarker. Through optimization of graphene chemical potential (and geometric parameters, the sensor achieves exceptional sensitivity of 1000 GHz/RIU with a quality factor ranging from 7.849 to 8.000. The integration of machine learning algorithms, including an ensemble of Random Forest, Support Vector Machine, and Neural Network models, significantly enhances analytical capabilities with 98.7 % prediction accuracy and 2.3 μg/mL RMSE. The ML framework incorporates advanced spectral pre-processing with 95 % noise reduction, automated extraction of 127 spectral features, and real-time processing capabilities with sub-second response times (0.12 s). Electric field distribution analysis reveals optimal resonance at 0.68 THz with maximum field confinement, while the sensor demonstrates robust performance across varying incidence angles. The proposed system offers superior detection limits, high selectivity, and exceptional reliability with 95.3 % average prediction confidence, making it highly suitable for point-of-care diagnostics, nutraceutical quality control, and personalized health monitoring applications.