Support Vector Regression for Gain and S11 Prediction: A Low-Complexity Solution for Antenna Design

This study explores the application of Support Vector Machines (SVM) for predicting key performance metrics, namely Gain and S11 (return loss), in microstrip patch antenna design. By leveraging the kernel-based approach, SVM effectively models the non-linear relationships between geometrical parameters and performance characteristics, delivering accurate predictions with reduced computational overhead. The model was trained on a dataset of geometrical parameters and corresponding performance metrics, achieving Mean Absolute Error (MAE) values of 0.0024 for Gain and 0.008 for S11 on a test set of 456 samples. Compared to Neural Networks (NNs), SVM exhibits faster training times, lower computational complexity, and reduced susceptibility to overfitting, making it more suitable for real-time design optimization tasks. While NNs can capture intricate non-linear dependencies more effectively, SVM balances prediction accuracy with computational efficiency, offering a scalable and interpretable solution for antenna optimization. These findings highlight the potential of SVM as a robust alternative to traditional electromagnetic simulations and more resource-intensive machine learning models, enabling rapid design iterations and performance evaluation in resource constrained environments.