A hybrid BOBE framework for scattering parameter prediction of wideband crescent slot microstrip antenna

Electromagnetic simulation of microstrip antennas for wide-band operation often requires extensive computational resources, which limits rapid design optimization for modern wireless communication systems. In this work, a compact wide-band crescent-slot microstrip antenna with a partial ground plane is designed to operate at 3.5 GHz and 7.0 GHz. The antenna achieves excellent impedance matching with return-loss values of − 28 dB and − 26 dB at the respective resonant frequencies while maintaining stable radiation characteristics across both bands.To minimize the computational load due to repeated full-wave simulations, a machine learning approach based on the Bayesian Optimized Blended Ensemble (BOBE) algorithm is proposed to predict the reflection coefficient ( S 11 ) parameter. The proposed algorithm combines different regression models namely Random Forest, XGBoost, CatBoost and Extra Trees, and uses Bayesian optimization to blend their predictions. In the proposed approach, the non-linear relationships between the antenna geometry and the EM characteristics are well captured, with an accuracy of R 2 = 0 . 991 , RMSE = 0 . 252 , MAE = 0 . 24 , and MAPE = 2 . 05 % .In comparison to other conventional simulation-based methods that take up to several hours of computational time, the proposed model achieves the task of predicting S 11 responses within milliseconds while maintaining high accuracy. The integration of the optimized design of the antenna with the proposed data-driven ensemble learning framework provides an efficient methodology to accelerate the design of antennas for 5G and next-generation wireless communication systems.