Large-Scale IP Network Flow-Level Abnormal Behavior Trends Analysis Using Combined Anomaly Grey Forecasting Algorithm

Abstract

Network traffic analysis has received academic attention in recent years; nevertheless, few have investigated the case that the network traffic data collected may include missing values and sufficient network traffic data may not be acquired for privacy protection or the limitation of network storage equipment capacity. This paper investigates flow-level abnormal behavior trends prediction in large-scale IP networks by the means of analyzing small samples taken from massive data information. We propose an algorithm called Independent-Components-Analysis-Anomaly-Grey-Forecasting (ICA-AGF) and conduct experiments to evaluate the algorithm with Abilene network Netflow data.