ANGEL - Machine Learning Classification

The Automated Network Games Enhancement Layer (ANGEL) project aims to leverage machine learning techniques to automate the classification and isolation of interactive (e.g. games, voice over IP) and noninteractive (e.g. web) traffic. This information is then used to dynamically reconfigure the network to improve the Quality of Service provided to the current interactive traffic flows and subsequently deliver improved performance to the end users. In this document we first describe the current solutions for network traffic classification and their shortfalls. Then we present the novel approach of classifying network traffic flows based on machine learning techniques and statistical flow attributes such as packet length and interarrival time statistics. We evaluate if a machine learning based approach meets the accuracy and performance requirements of the ANGEL architecture and select appropriate machine learning algorithms. The evaluation is based on results obtained for real game traffic captured in the network.