Analysis on Performance Comparison Between Empirical and Machine Learning Path Loss Models

Abstract

Deployment of mobile networks requires accurate Link Budget to determine the heights of antennas, transmit power, and cell radius. An efficient Link Budget is only possible when path loss has been determined accurately. Empirical models are widely used for prediction of path loss but their accuracy tends to be low. Hence, they are sometimes tuned to fit the environment under consideration if path loss measurements are available. Machine Learning (ML) Models have evolved with time and their performances have surpassed that of Empirical Models. Although the accuracy of Tuned Empirical Models is better than the existing Empirical Models, performance comparison between the accuracy of Tuned Empirical Models and ML models is not reported in literature. This paper identified researches in which Comparisons were made between Empirical models and Tuned Empirical Models and determined an average performance improvement of the tuning process as 56.12%. It also identified researches in which the performances of ML and existing Empirical Models were observed. Predictions were made on how the accuracy of these Empirical Models that have been compared with ML Models will be in case they have been tuned. This was achieved using the average improvement in performance introduced by tuning process (56.12%) computed earlier. Results showed that the performance of the Tuned Empirical Models can be as good or sometimes better than that of ML Models. It is therefore recommended that comparisons should be made between Tuned Empirical Models and ML Models for determination of a suitable model especially when considering the fact that the Tuned Empirical Models are Glass Box Models such that their explanation is clear.