Forward-inverse-hybrid modeling of microstrip antennas using decision tree-based machine learning algorithms for space communication

• Published in: International journal of electronics and communications Vol. 191; p. 155662
• Main Authors: Kumar, Anjani, Khan, Taimoor, Sarkar, Debanjali
• Format: Journal Article
• Language: English
• Published: Elsevier GmbH 01.02.2025
• Subjects: AdaBoost; And random forest regression; Binary decision tree; Machine learning; MIMO microstrip antenna; Satellite environment; Binary decision tree; And random forest regression; AdaBoost; Satellite environment; MIMO microstrip antenna; Machine learning
• ISSN: 1434-8411
• DOI: 10.1016/j.aeue.2025.155662

This article presents a machine learning (ML) framework for forward, inverse, and hybrid modeling in electromagnetic problems, focusing on single-, dual-, and four-port notched microstrip antennas used in satellite environments. These antennas exhibit maximum notched frequencies of 4.2 GHz, 6 GHz, and 8.2 GHz for the single-port, dual-port, and four-port configurations, respectively, making them well-suited for space communication in low Earth orbit and medium Earth orbit. The study explores the application of binary decision trees, AdaBoost, and random forest regression (RFR) for these modeling tasks. The single-, dual-, and four-port microstrip antennas feature three, four, and two notched bands, respectively. The accuracy of various ML techniques is validated by comparing their predicted results with those obtained from a high-frequency structure simulator (HFSS). Machine learning techniques are used to predict the mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and coefficient of determination for the geometrical and electrical parameters of these antennas. Among the tested methods, AdaBoost algorithms demonstrated significantly superior performance compared to binary decision trees and random forest regression.