Unsupervised Anomaly Detection for Communication Networks: An Autoencoder Approach

Abstract

Communication networks are complex systems consisting of many components each producing a multitude of system metrics that can be monitored in real-time. Anomaly Detection (AD) allows to detect deviant behavior in these system metrics. However, in communication networks, large amounts of domain knowledge and huge manual efforts are required to efficiently monitor these complex systems. In this paper, we describe how AutoEncoders (AE) can elevate the manual effort for unsupervised AD in communication networks. We show that AE can be applied, without domain knowledge or manual effort and evaluate different types of AE architectures and how they perform on a variety of anomaly types found in communication networks.