Machine learning-assisted multi-band metamaterial absorber for terahertz sensing and gas detection

Abstract A multi-band metamaterial absorber operating in the terahertz region is designed using two concentric octagonal ring resonators with peripheral conductive structures and a metallic board separated by a dielectric spacer. The simulated results demonstrate five distinctive absorption modes at 2.64, 4.21, 5.43, 7.9, and 8.5 THz, with absorption rates of 96.8%, 99.5%, 92.87%, 99.6%, and 97.5%, respectively. To optimize absorption performance, machine learning models—including CatBoost, ExtraTree, and KNN—were employed to predict and refine the influence of geometric parameters on multi-band absorption. Among these, the KNN model exhibited the best performance, achieving an R-squared value of 0.9915, with an RMSE of 0.0220 and an MAE of 0.0090, indicating superior prediction accuracy. Furthermore, electric field, magnetic field, and surface current distributions were analyzed to understand the physical mechanisms behind the absorption peaks A sensor based on this design is proposed for harmful gas detection, demonstrated for Methane and Chloroform detection. This study enhances the adaptability of the metamaterial absorber for terahertz sensing, imaging, and communication applications.