A Comprehensive Survey of Machine Learning Based Pathloss Models for Wireless Communication Systems

Abstract

With the advent of 5G networks and advancements in wireless communication, the need for adaptive and robust path loss prediction models has become increasingly critical. Although traditional empirical and deterministic models have served as foundational tools, they often fail to effectively capture the intricacies of modern wireless propagation environments. This shortfall has spurred a transition toward machine learning (ML) and deep learning (DL) techniques, which demonstrate greater adaptability across varying scenarios and frequency ranges.

This paper provides a comprehensive review of existing path loss prediction methods, focusing on the potential of ML and DL approaches as viable alternatives to conventional models. It explores a range of methodologies, from shallow algorithms such as Random Forest, Gradient Boosting, Support Vector Regression, and Artificial Neural Networks to advanced approaches like Convolutional Neural Networks, hybrid deep learning frameworks, and Adaptive Neuro-Fuzzy Inference Systems. By analyzing their performance, advantages, and limitations, the study identifies key trends and uncovers research gaps in this evolving field.

The findings aim to inform future research efforts aimed at designing path loss prediction models that are more accurate, efficient, and capable of adapting to the demands of next-generation wireless communication systems.