Mining Graph-Fourier Transform Time Series for Anomaly Detection of Internet Traffic at Core and Metro Networks

Abstract

This paper proposes a framework to analyse traffic-data processes on a long-haul backbone infrastructure network providing internet services at a national level. This type of network requires low latency and fast speed, which means there is a large demand for research focusing on near real-time decision-making and resilience assessment. To this aim, this paper proposes two innovative, complementary procedures: a multi-view approach for the topology analysis of a backbone network at a static level and a time-series mining approach of the graph signal for modelling the traffic dynamics. The combined framework provides a deeper understanding of a backbone network than classical models, allowing for backbone network optimisation operations and management at near real time. The applications of this methodology to the backbone infrastructure of one of the main internet service providers in the UK shows increased accuracy and computational efficiency on the detection of where and when anomalies and irregular patterns occur in the network signal