Machine Learning for Microstrip Patch Antenna Design: Observations and Recommendations

In this paper, the relationship between the configuration parameters of microstrip patch antenna and its resonant frequency is analyzed from the machine learning aspect. By training two datasets with 258 and 4209 antenna configurations, the linear regression model shows that the mean value of resonant frequency accuracy prediction $\boldsymbol{R^{2}}$ score achieves 0.582 and 0.647 under 3 times of repeated 5-fold cross-validations for the operational frequency range from 3 to 15 GHz. Larger datasets with higher prediction accuracy would make the machine-learning-based model a potential method in antenna design without the need for massive dimensional tuning in full-wave simulators for this type of patch antenna.