Developing an ensemble learning model for slicing 5G networks

Contemporary mobile communication systems require a flexible infrastructure that can accommodate a range of services that require different performance characteristics, so network slicing has become a necessary mechanism for generating multiple isolated virtual networks, built on top of a single physical infrastructure, to provide services best suitable for the application performance requirements. In this paper, an ensemble learning model for network slice allocation in 5G networks is presented based on machine learning techniques. The proposed model combines ensemble learning techniques such as Random Forest (RF), CatBoost (CT), and AdaBoost (ADB), with Logistic Regression (LR) as the final decision-making stage. The proposed ensemble model demonstrated performance with an accuracy of 98.32%, which is over 2% better than the best single model (CatBoost at 96.23%). The proposed ensemble model consistently outperformed Random Forest and AdaBoost by more than 4% and more than 5% respectively. In addition, the proposed ensemble model was balanced and across key 5G slice types (eMBB, mMTC, URLLC) supported the diverse service requirements.