Machine learning-based prediction of design parameters for miniaturized textile antennas in the sub-6 GHz spectrum

This paper presents a novel antenna miniaturization approach for wearable applications, based on resonant current path length manipulation using T-shaped slots without compromising performance. A square microstrip textile antenna, initially designed for 5.8 GHz operation, undergoes modifications with T-shaped slots of varying lengths (11, 22, and 28 mm). These slots extend the effective current path, thereby reducing the resonance frequency to 5.2, 3.5, and 2.4 GHz, respectively, resulting in an 84% reduction in antenna size. The novelty of this work combining slot-based miniaturization with Machine Learning (ML) driven optimization. ML models, including Decision Tree (DT), Linear Regression (LR), Random Forest (RF), Gradient Boosted Tree (GBT), and Support Vector Machine (SVM), are evaluated using Root-Mean-Square Error (RMSE), Mean Absolute Error (MAE), and correlation values. Among the models, Decision Tree demonstrates superior accuracy with RMSE = 0.023, MAE = 0.019, and a correlation value of 0.998, demonstrating that ML can effectively support the design process, reduce iterative simulations. The combined slot-based miniaturization and ML-driven optimization framework provides a compact, high-performance solution for sub-6 GHz wearable antennas, offering both practical manufacturability and efficient design.