Explainable Machine Learning Model for Path Loss Prediction in 5G Networks

Abstract

<title>Abstract</title> This study presents an interpretable machine learning framework for accurate path loss prediction in IoT-enabled 5G networks using Support Vector Regression (SVR) combined with Shapley Additive Explanations (SHAP). Traditional empirical and heuristic path loss models often fail to capture the nonlinear and dynamic characteristics of modern wireless environments. To address this challenge, an SVR model with a Gaussian kernel was trained using an IoT dataset obtained from the Zenodo repository, after a data preprocessing workflow. Model performance was evaluated using mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). The SVR model demonstrated high predictive accuracy with RMSE = 0.414 dB, MAE = 0.096 dB, and R² = 0.999. This indicates its capability to represent complex signal propagation behavior effectively. SHAP analysis further revealed that received power (Prx) and received signal strength indicator (RSSI) accounted for over 90% of the prediction influence, establishing them as the key determinants of path loss. Spatial and environmental variables, including distance, wall obstruction, and frequency, showed minimal effect. The integration of SVR with SHAP improves both prediction accuracy and interpretability, supporting informed decision‑making in IoT‑driven 5G network planning.