Performance Evaluation of Machine-Learning Models for Self-Healing in 5G Networks

Abstract

The 5G self-organizing network is a viable solution to the problem of increasing user-connectivity, data rates, and network complexity. This paper proposes a SON solution that uses machine learning for anomaly detection in order to meet user demands. Three different supervised ML algorithms are used for anomaly detection to see which provides the most efficient and accurate results. The various algorithms used key performance indicators (KPIs) to determine whether a base station is healthy, congested, or failing. In order to achieve unbiased results, large datasets composed of multiple simulated network scenarios were preprocessed and partitioned for training and testing. The results show that state vector machine algorithm can accurately detect the status of a base station at exponentially lower processing times than the other ML algorithms. This algorithm was most efficient when larger datasets were used to create the model.