AI-driven self-optimisation of network control functions in multi-tenant beyond 5G networks

The integration of Artificial Intelligence (AI) in mobile networks driven by softwarisation and virtualisation offers innovative solutions for network control and network optimisation. This research focuses on a concrete case of self-optimisation network control in Beyond 5G Networks (B5G), investigating the use of different network control functions for optimal implementation of network rules. This thesis proposes an AI-driven approach to select the most appropriate network control function technology (e.g., Linux Traffic Control, iptables, Open Virtual Switch) based on specific network conditions and external factors. For this purpose, this research work first analyses key metrics for network control and creates a comprehensive training dataset. To make the creation of this dataset possible, two software components have been developed. A network monitoring component and a network data extractor capable of gathering more than 100 metrics in less than 1 second. The networking dataset has been extracted from a lab-based B5G network by running four different use cases. The resulting dataset, called the IW-IB-5GNET dataset, consisting of 7,708 labelled instances, was labelled using a newly developed labelling tool, NetLabeller. Labels represent the optimal network control technology for each scenario deployed£. After the labelling process, various data augmentation techniques, including GAN-based algorithms, were performed to enrich the dataset. Finally, a Random Forest classifier has been selected as a result of the empirical comparison of different supervised ML algorithms. The Random Forest outperforms the rest of the classifiers, obtaining a total of 90.83% F1-score and an accuracy of 97.10%. This research work and its contributions collectively prepare the way for advanced, autonomous B5Gnetworks by optimising network policies using AI algorithms.