Intelligent Network Data Classification

Abstract

In present, lot of attention has been given to network data classi cation. Applications of this area varying from maintaining a good quality of service to analyse personal charac- teristics of users of network. With assist of intelligent systems, Internet service providers can use network classi ers to allocate fair bandwidth to their users and network network administrators to detect anomalies as well. Network classi cation domain has strengthen its power within past decade with aid of lot of other areas. At the beginning network classi ers use traditional classi cation techniques such as port based classi cation and payload based classi cation. With the development of other areas in networking domain such as protocols and applications, traditional meth- ods have become less usable. This makes researchers to focus on using machine learning, statistical methods and heuristics base methods for network classi cation. World wide web is a big graph made up with URLs and IP addresses as nodes and hyperlinks connecting two IP addreses (or URLs) act as edges. Also this graph which is made out from world wide web is dynamic graph where each second thousands of new hyperlinks keep adding. For a given entity only a portion of this large graph, which made out of URLs and IP addresses that entity browse, is visible. This is true for local area networks as well. Since the world wide web is a graph and graph theories can apply on it, this research focuses on using graph theories and graph partition methods to partition the collected network tra c. Hence the proposed method can be use as a tra c classi cation method.