A Fusion Framework Combining DNN and Random Forest for 5G Millimeter-Wave Antenna Design and Optimization

This article presents a fusion framework combining deep neural network (DNN) and random forest (RF) to enhance the design of a compact C-shaped patch antenna (CSPA) for 5G millimeter-wave applications. The proposed framework leverages the DNN’s ability to learn complex, nonlinear relationships and the RF’s robustness to noise, blending predictions to accurately estimate critical antenna metrics such as S11 and gain in dBi. This approach significantly reduces the computational cost of full-wave simulations, making it a powerful tool for rapid antenna prototyping and performance enhancement in modern wireless communication systems. To generate the database for training and testing the model, CSPAs with different geometrical and electrical parameters are simulated in terms of the resonant frequency using HFSS. The antenna design, featuring a C-shaped patch on a grounded substrate with an overall area of $12\times 6\times 0.8$ mm, operates across the 23.1–48.9-GHz band, achieving a peak gain of 2.22 dBi. In addition, this article also provides a comparative analysis against state-of-the-art machine learning (ML) and DNN models, which demonstrates that the proposed DNN + RF framework offers superior accuracy, faster convergence, and robust performance, making it a promising solution for next-generation antenna design in 5G and beyond. It reduces the mean square error (mse) to 0.0021 and the mean absolute error (MAE) to 0.045, with an average relative error dropping below 1.5% after 140 iterations. The blended predictions show enhanced accuracy, as indicated by the scatter plots aligning closely with the ideal prediction line. This data-driven approach accelerates antenna optimization, providing a robust framework for high-frequency wireless systems.