Characterization of Network Traffic Data: A Data Preprocessing and Data Mining Application

摘要

Large amount of traffic data are transmitted during day-to-day operation of wide area networks. Due to the increment of diversity in network applications, its traffic features have substantially changed. Data complexity and its diversity have been rapidly expanding with the changing nature of network applications. In addition, bandwith and speed of network have increased rapidly as compared to the past. Therefore, it is a necessity to characterize the changing network traffic data to understand network behavior. The aim of this research is to understand the data nature and to find useful and interesting knowledge from the network traffic traces which contains IP protocol packets. We analyze the traffic trace of 21 April 2012 on a 150 Mbps transpacific link between US and Japan from the MAWI Working Group traffic archive. This data contain lots of useful and important information which is hidden and not directly accessible. In this research, firstly, anomaly detection analysis and Kohonen Networks are applied to reduce the data matrix. Then, we generate a CART decision tree model to mine traffic data. The decision tree method is successfully applied in network traffic analysis. The results show that the proposed method has substantially good performance.