Machine Learning-Based Path Loss Models: Towards a Unified Methodology

Abstract:

Machine Learning (ML) enhances wireless signal propagation models by leveraging diverse data sources and improving prediction accuracy. However, many existing models struggle to generalize due to inappropriate features and limited evaluation methods. This work reviews ML techniques for path loss prediction across various wireless technologies, including 5G and beyond, in different environments. It also introduces a structured methodology to improve model performance by selecting relevant features, using advanced evaluation techniques, and applying spatial cross-validation for better generalization. The proposed approach incorporates key environmental and system parameters while avoiding overfitting from geographic coordinates with the training area separated from the test area. ML models should also be compared to traditional models in wireless communications to confirm accuracy and computational efficiency. Further research is needed to experimentally validate this methodology as a standard framework for ML-based path loss models. In the future, this framework could help define standardized criteria for ML-based path loss modeling in telecommunications.

DOI: https://doi.org/10.1109/ICETCI67340.2025.11258025