Accelerating Compact UWB Fractal Antenna Design Using Interpretable Machine Learning with Forward and Inverse Prediction

Designing compact ultra-wideband (UWB) antennas with wide impedance bandwidth remains a significant challenge because antenna performance strongly depends on multiple interdependent geometrical parameters, often requiring extensive electromagnetic (EM) simulations, repeated optimization, and expert-driven manual tuning. Although recent machine learning (ML) approaches have attempted to automate antenna design, many rely on computationally intensive black-box models that offer limited interpretability and mainly support forward-only prediction. To address these challenges, this work presents a compact fractal antenna integrated with a lightweight and interpretable ML-assisted framework for both forward and inverse prediction. The proposed antenna employs irregular fractal slots and a defected ground structure to enhance the effective electrical length and achieve wideband performance over the 2.5–15 GHz frequency range while maintaining compact size. The optimized antenna achieves measured fractional bandwidths of 76.38% and 40.28% in two operating bands (2.8–6.26 GHz and 9.97–15 GHz). A Decision Tree Regression (DTR)-based framework is developed to predict resonance frequencies, bandwidths, and antenna dimensions with up to 99% prediction accuracy using only 162 CST-generated samples. The inverse prediction framework further enables direct estimation of antenna dimensions from desired performance targets, significantly reducing iterative optimization complexity and simulation time. The measured and simulated results show good agreement, confirming the practical feasibility and efficiency of the proposed ML-assisted compact antenna design methodology for future wireless and array-integrated communication systems.