Performance Analysis of Regular Polygonal Patch Antennas for Band Classification Using Machine Learning

This study evaluates the performance of machine learning models in classifying the number of frequency bands in microstrip patch antennas based on geometric parameters. The patches range from triangular to circular, with key input features being cut width, feed width, and patch size. The dataset, derived from High-Frequency Structure Simulator (HFSS) simulations, ensures precise antenna parameter measurements across a frequency range of 2 to 7 GHz. Four machine learning models—Support Vector Classifier (SVC), Random Forest, Extra Trees, and KNearest Neighbors (KNN)—were analyzed for their classification accuracy across different patch geometries. Results indicate that ensemble models such as Extra Trees and Random Forest outperform other classifiers, achieving the highest accuracy in predicting up to three bands. Additionally, hexagonal, octagonal and circular patches exhibited superior classification performance compared to triangular and pentagonal patches.