Symbolic Regression for Large-Scale Propagation Modeling in the Atlantic Forest

This paper investigates the use of symbolic regression to derive analytical models for large-scale propagation in the Atlantic Forest environment using LoRa measurements. Unlike traditional empirical or machine learning approaches, symbolic regression provides concise and interpretable equations. Models are obtained using distance and spreading factor as inputs. Despite the small dataset, the derived formulas achieved a Pearson correlation coefficient above 0.98 with the measured data, indicating strong predictive capability. Sensitivity analysis reveals physically meaningful trends, such as increasing path loss with distance and improved received power with higher spreading factors. The results highlight the potential of symbolic regression as a lightweight and interpretable alternative for propagation modeling.