Detecting Network Degradation Using Machine Learning Predicting abnormal network behavior with anomaly detection

Abstract

Due to the prohibitive cost of downtime in large complex systems, it is important to reduce or entirely eliminate any downtime that might happen as a result of degradation in system quality. This thesis paper presents a newly developed model for network congestion detection in large scale networks, using Gaussian Mixture Model and Isolation Forest algorithms to improve early detection of failure in the Ericsson EBM data. The results from the evaluation suggests that the developed method is more robust and precise in finding anomalies, compared to the current method used by Ericsson.