Tunable dual-band THz metamaterial absorber with regression-learning-enabled numerical redesign

Terahertz (THz) metamaterial absorbers have garnered considerable interest for applications in sensing, detection, and stealth technologies due to their capacity to achieve engineered absorption at subwavelength scales. This study introduces a high-performance THz metamaterial absorber featuring integrated circular split-ring resonators patterned in copper over a lead glass substrate. The design achieves dual strong absorption peaks at approximately 1.44 THz and 4.24 THz, each with absorption amplitudes nearing 90%. To explore the influence of geometric parameters on absorption characteristics and facilitate predictive modeling, full-wave electromagnetic simulations were performed by systematically varying key design variables. The resulting dataset comprises geometric parameters as input features and corresponding resonant frequencies and absorption amplitudes as target outputs. Two machine learning strategies were employed: forward design, predicting spectral response from geometry, and inverse design, estimating geometry from desired frequency peaks. Multiple linear and tree based regression models, including linear, lasso, random forest, decision trees, gradient boosting models were trained and evaluated using standard error metrics such as RMSE, MAE, and $$R^2$$ etc. A physics-informed confirmation step was carried out by re-simulating specific regression-predicted designs in a full-wave solver and comparing the resulting resonance frequencies and absorption levels with the intended targets, producing low percentage errors. Because the inverse design approach allows for the immediate retrieval of geometry corresponding to user-specified resonance frequencies and absorption amplitudes, facilitating quick spectrum-to-structure design, this validation further illustrates the tunability of the suggested absorber. All things considered, the suggested data-driven methodology reduces dependency on computationally costly repetitive simulations by facilitating effective optimization and quick prototyping of THz absorbers.