CTFI: Adaptive pattern analysis of internet traffic using time series compressibility

Nowadays, network management is a more challenging task, since the Internet has been changing continuously. Although new problems, for example, excessive bandwidth consumption, continue to appear due to the development of new applications, high-resolution data such as packet or flow level data is often unsuitable for privacy protection reasons. For networks with these kinds of problems, we propose a novel traffic classification algorithm, which has the capability to learn a variety of traffic patterns and to track traffic movement. The proposed method can classify aggregate traffic automatically without a priori modeling by a label series pattern classification based on compressibility. The proposed technique enables the automatic separation of traffic from contracts or assumptions and realizes the isolation of unknown traffic almost without human intervention.