Machine Learning Driven Optimization of a Triple-Band Terahertz Metamaterial Perfect Absorber for Biomedical Applications and Biomarker Detection

The integration of machine learning into terahertz (THz) metamaterial design has opened new frontiers in high-sensitivity, label-free biomedical sensing. This study presents a triple-band perfect metamaterial absorber whose microstructural geometry was rigorously optimized using a synergistic framework combining finite-difference time-domain (FDTD) simulations, genetic algorithm (GA), and machine learning (ML) approach, achieving near-unity absorption of 99.92%, 99.97%, and 99.58% at three distinct THz resonances. The machine learning driven optimization yields a dramatic performance boost over prior multi-band absorbers. Random Forest Regression outperformed other models, achieving the MSE (0.000405), lowest RMSE (0.0201) and highest R² (0.9947), indicating superior accuracy and generalization. The optimized absorber exhibits sharp, high-Q resonances with ultra-narrow linewidths, enabling precise identification of subtle spectral shifts caused by cancer biomarkers, viral particles, parasites, glucose concentrations, and tissue-specific signatures. A peak sensitivity of 10,142.86 GHz/RIU was achieved, along with polarization insensitivity across 0° to 90°, and angle-tolerant performance attributes that collectively underscore its clinical viability. This triple-band THz metamaterial biosensor facilitates simultaneous, non-invasive detection of multiple analytes within complex biological environments, providing a compelling platform for next-generation biomedical diagnostics.