Comparison of Machine Learning Algorithms for Application in Antenna Design

Modern antennas have complex structures, and the design of these devices is a challenging task. The antenna optimization process uses electromagnetic simulations as an objective function, which have a high computational cost. Surrogate models are methods that can be used to increase computational efficiency by training machine learning techniques. This work investigates eight machine learning techniques to find which one is more suitable to be used as a surrogate model in the design of antennas. We propose a methodology for comparing and analyzing the techniques using the RMSE as the metric. For the case study, we present the design of a Quasi-Yagi antenna for operating at three resonance frequencies. The results demonstrate that the Gaussian Process model obtained the best performance, achieving an RMSE value of 1.251 in the case study.