Machine Learning-Based Excess Path Loss Modeling for S-Band Terrestrial Radio Propagation

Accurate radio propagation modeling is essential for the design and planning of modern terrestrial and satelliteassisted communication systems operating in the S-band. Classical empirical models such as the COST-231 Hata model often fail to account for environment-dependent excess losses caused by atmospheric and local propagation effects, leading to significant prediction errors. In this work, a data-driven framework is proposed to model excess path loss beyond conventional propagation models using supervised machine learning. A synthetic propagation dataset is generated by combining free-space path loss, COST-231 predictions, and additional attenuation due to rain and atmospheric parameters. Excess loss is defined as the deviation between the total simulated path loss and the COST-231 baseline. Multiple regression models are evaluated, including linear regression and random forest regression. Experimental results demonstrate that the random forest model accurately captures the nonlinear behavior of excess loss, achieving a root mean square error of 0.035 dB, significantly outperforming traditional empirical approaches. The proposed framework highlights the potential of machine learning techniques to enhance classical propagation models for improved radio link planning.