Explainable Machine Learning for Path Loss Modeling in mmWave and Sub-THz Communication Systems

Machine learning has emerged as a powerful tool for large-scale path loss modeling at millimeter-wave (mmWave) and sub-terahertz (sub-THz) frequencies. While recent studies demonstrate improved prediction accuracy, most learning based models operate as black boxes, limiting their adoption in deployment-critical wireless system design and stan dardization activities. This paper addresses this limitation by proposing an explainable machine learning framework for path loss modeling that emphasizes interpretability, physical consis tency, and engineering trust. Using multi-band mmWave and sub-THz measurement datasets, we analyze how key propa gation features—including transmitter–receiver separation dis tance, carrier frequency, antenna height, and line-of-sight condi tions—influence model predictions. Feature attribution and sen sitivity analyses are employed to quantify the relative importance of these parameters and to assess whether learned relationships align with established propagation theory. The results demon strate that the proposed models capture physically meaningful monotonic trends with respect to distance and frequency, while avoiding non-physical artifacts commonly observed in purely data-driven approaches. Rather than prioritizing marginal accu racy gains, this work highlights the tradeoff between predictive performance and model transparency, showing that interpretable models can achieve competitive accuracy while offering substan tially improved explainability. The presented framework provides actionable insights for network planning, site-specific modeling, and data-assisted calibration of analytical channel models. These findings support the role of explainable machine learning as a key enabler for trustworthy and deployable path loss modeling in future mmWave and sub-THz wireless communication systems.