Journal Paper

This paper presents a novel Explainable Artificial Intelligence (XAI) framework for urban
radio propagation modeling based on Symbolic Regression (SR). Unlike conventional empirical models
and black-box machine-learning approaches, the proposed method autonomously discovers compact
analytical expressions for path loss directly from measurement data without imposing predefined functional
forms. Real-world propagation measurements collected at 735 MHz, 2.54 GHz, and 3.5 GHz in urban
environments are used to derive and evaluate the proposed model. To the best of the authors’ knowledge,
this is the first study to demonstrate the application of SR to multi-frequency sub-6 GHz urban path-loss
modeling using real measurement campaigns. The resulting SR-derived models provide prediction accuracy
comparable to optimized empirical formulations while maintaining full physical interpretability. To assess
robustness and generalization capability, the proposed model is further validated using an independent
measurement dataset collected in a different urban environment, transmitter configuration, and operating
frequency. In addition, a comparison with a Random Forest regressor highlights the trade-off between
prediction accuracy, interpretability, and generalization. The results demonstrate that SR can generate
physically meaningful propagation models with competitive accuracy and improved robustness under
previously unseen conditions, effectively bridging empirical modeling, physical knowledge, and machine
learning for next-generation wireless networks, digital-twin applications, and future 6G systems.