Performance investigation of machine learning methods in 5G and beyond network slicing

Thanks to the rapid developments in the communication sector in recent years, new applications, and opportunities have emerged. These applications have created the need for low latency, high data rate, high reliability, and security. New proposals had to be given for these needs that the fourth generation (4G) communication systems could not meet, and fifth generation (5G) communication systems emerged. For example, while software defined networking (SDN), and network function virtualization (NFV) technologies are used for programmable structures in 4G systems, it is planned to use network slicing methods in addition to these two technologies in future 5G systems. In this study, 5G Network Slicing, Data Rate Management, and User Handover Mechanisms were created with machine learning models, and additional mechanisms. In order to test these mechanisms, a realistic simulation environment has been developed. 3rd Generation Partnership Project (3GPP) compatible users and base stations are placed in this simulation environment. Five different machine learning models were used in the study, and four different models for network slicing, and a classification mechanism based on the majority of these models' decisions were implemented. In order to test the performance of the created machine learning models in realistic scenarios, user data with margin of error was created. With this user data, classification has been carried out with data with error margin that can be experienced in real situations. Classifications were performed between the enhanced mobile broadband (eMBB), ultra-reliable low latency communications (URLLC), massive internet of things (MIoT), and vehicle to everything (V2X) network slices that are proposed by 3GPP. The implemented mechanisms were tested in different scenarios, and their performances were compared.