Near Real Time Online Flow-Based Internet Traffic Classification Using Machine Learning (C4.5)

Offering reliable novel service in modern heterogeneous networks is a keychallenge and an important prospective income source for many networkoperators and providers. Providing reliable future service in a cost effectivescalable manner requires efficient use of networking and computing resources.This can be done by making the network more self enabled, i.e. making itcapable of making distributed local decisions regarding the utilization of theavailable resources. However such decisions must be correlated in order toachieve the global overall goal (maximizing the performance and minimizing thecost)Since network administrators are always worried about making fast decisions tomonitor and regulate the Internet traffic, a novel approach for online flow-basednetwork traffic classification is proposed. This proposal is based on Machinelearning algorithm C4.5 and a custom built network traffic data set captured froma university campus environment. Furthermore the aim of this effort is to build acomplete online flow based traffic classification and control system.Validation on the proposed system is done from accuracy and time points ofviews. Firstly, an offline training and testing data sets are applied to Weka’s C4.5and our system. And their corresponding accuracy has been compared. Ourexperimental results show that the accuracy is the exactly the same. Secondly,the received UDP NetFlow packets have been send to our system and to a basicpacket sniffing program and the number of NetFlow packets has been counted ineach. The comparison result show that no packet overwriting due to racecondition.