Application of Machine Learning Algorithms to Path Loss Modeling: A Review

The demand for high-speed internet services is increasing due to emerging needs such as e-commerce, e-health, education, and other high-technology applications. Wireless communication networks have now become a necessity, especially with the introduction of the 5G networks which have the potential to provide extraordinary data rates with extremely low latency. The deployment and operation of 5G and beyond networks in built-up environments would require a complex and reliable radio propagation model that guides network engineers to achieve effective coverage estimation and appropriate base station placements. The inefficiency, and sometimes inconsistencies of deterministic and empirical path loss models necessitated the need to integrate machine learning models. Recently, different machine learning-based pathloss models have been developed to overcome drawbacks associated with conventional pathloss models due to their significant learning and prediction abilities. This paper aims to review path loss models relative to machine learning-based algorithms with a focus on models developed in the last 21 years (2000 to 2021) to study their network parameters and architectures, designs, and applicability, and proffer further research directions.