Model Selection for Path Loss Prediction of Ultra High Frequency Terrestrial Television

• Ayodele S. OLUWOLE

  Author

• Olumayowa A. OJO

  Author

• Olaitan AKINSANMI

  Author

Precise prediction of path loss is crucial for the reliable operation and optimal coverage of terrestrial Ultra High Frequency (UHF) television networks. Traditional models tend to have limited accuracy in complex propagation environments due to the fixed parameter assumptions and the lack of consideration of the nonlinear interactions among terrain, vegetation and antenna characteristics. This paper investigates the applicability of different machine learning models for the prediction of path loss in terrestrial UHF television networks. Received signal strength was first measured in the field at selected locations along five different routes in Ekiti State, Nigeria. The measured dataset was subsequently preprocessed and prepared for machine learning analysis. Five supervised machine learning algorithms were implemented and trained on the dataset, such as Linear Regression, K-Nearest Neighbors Regression, Random Forest Regression, Decision Tree Regression and Extra Tree Regression. Lastly, , the performance of the models was assessed using root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). The results showed that KNN Regression had the lowest MAE (1.721), MSE (4.667) and RMSE (2.160) with an accuracy of 94.80%. Random Forest Regression was next with a slightly higher MAE (1.733), MSE (6.775) and RMSE (2.603), but scored the highest accuracy (94.89%) as a balanced and reliable performance overall. The results validate random forest as optimal and reliable path loss prediction model in UHF terrestrial broadcast networks. Carefully selection of machine learning models helps achieve precise estimation of path loss.

• 29_Oluwole et al. (2026)

Copyright (c) 2026 Ayodele S. OLUWOLE, Olumayowa A. OJO, Olaitan AKINSANMI (Author)

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.