PREDICTING ANTENNA S11 USING GAUSSIAN PROCESS REGRESSION, EXTREME GRADIENT BOOSTING, AND MULTILAYER PERCEPTRON

Abstract

Machine Learning surrogates can accelerate antenna design by predicting key RF responses without repeated and cumbersome EM simulations. In this study, three different machine learning models were used to predict the S11 of Microstrip Patch Antenna. These algorithms belong to different families: Gaussian Process Regression (GPR), a kernel Bayesian non-parameteric regressor; Xtreme Gradient Boosting(XGB), a decision tree-based algorithm well-suited for non-linear tabular data; and a Multi-Layer Perceptron (MLP), a feed-forward neural network that learns nonlinear feature mappings. The algorithms were compared based on performance metrics such as MSE, RMSE, MAE, and R2 score. Amongst all three, XGboost stands out with an R² score of 0.9687.