Performance Analysis of Machine Learning Models for Predicting Return Loss in 5G Microstrip Patch Array Antenna Design

This work explores the integration of machine learning models to predict the return loss of a microstrip antenna operating at 3.5 GHz in the sub-6 GHz 5G frequency band. 11 (eleven) design parameters are varied to create a dataset of 380 samples. Eight regression models are used including Random Forest Regression, Decision Tree Regression, linear regression, and Extreme Gradient Boosting. Hyperparameter tuning is conducted via grid search and randomized search with cross-validation. Seven performance metrics, Mean square error, Mean absolute error, R-square, and Variance score, are used to evaluate the model. Random forest regression provides the best result with an MAE of 9.59%, R-square of 90.59%, and variance of 93.49%. This study also utilizes residual and feature importance analysis to understand the model performance better.