Data-driven approaches for predictive modeling of N77/N78 band 5G microstrip patch array antenna using curve fitting and supervised regression

This paper deals with the design of a single-element and [Formula: see text] bandwidth microstrip patch array antennas for N77/N78 band applications. The presented antennas are designed on a low-cost FR4 epoxy substrate with a dielectric constant of 4.3 and a thickness of 1.6mm while considering manufacturability issues. In order to evaluate the antenna's characteristics, an equivalent circuit model of the proposed microstrip patch array will be built in Keysight ADS, and the reflection coefficient will be compared with the simulated result obtained with CST Studio Suite. An implemented prototype is also measured for the validity of the simulated results, and good agreement between the measurement and simulation is obtained. Besides the classical electromagnetic simulation, supervised machine learning (ML) algorithms are used to estimate the main antenna parameters, such as the bandwidth and the center resonant frequency of the investigated antenna. 203 simulation-based data samples are generated by using CST Microwave Studio and employed to train five regression models, including GB regression, ET regression, DT regression, RF regression, and XGB regression. The models are evaluated in terms of variance score, coefficient of determination (R[Formula: see text]), mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE). Among all tested models, the Random Forest regression model produces the best results in terms of error and accuracy for predicting both bandwidth and center resonant frequency. The proposed work concludes as an effective antenna design method through integration of full-wave simulation, measurement, equivalent circuit modeling, and machine learning-based prediction for robust N77/N78 band implementation.