Machine-learning-based prediction of return loss for coaxial probe-fed microstrip patch antennas: a comparative study of SVR, ANN, XGBoost, and Random Forest

This study explores machine learning (ML) techniques to predict the return loss (S11) of a coaxial probe-fed microstrip patch antenna operating in the 2–5 GHz range. This work mainly focused on training the ML models namely SVR, ANN, XGBoost, and Random Forest (RF) on a dataset generated via Ansys-HFSS parametric sweeps, varying patch dimensions, coaxial feed radius, and substrate height. Among the models, Random Forest achieved the highest accuracy (R² > 0. 9823), outperforming ANN (~0. 9271), XGBoost (~0. 9659), and SVR (~0. 9301). The results demonstrate that ML models can accurately predict S11, reducing reliance on full-wave simulations. This approach enables fast antenna optimization by predicting it’s performance, improving design efficiency for wireless applications.