ODC: a method for online detecting ＆ classifying network-wide traffic anomalies

Abstract

A method for online detecting ＆ classifying traffic anomalies（ODC for short） from a network-wide angle of view was put forward.This method constructed traffic matrix with a metric of traffic feature entropy incrementally,de-tected traffic anomalies online using incremental principal component analysis,and then classified traffic anomalies online using incremental k-means,from which network operators could benefit for taking corresponding countermeasures.Theoretical analysis and experiment analysis show that the method has lower storage and less computing time complexity,which could satisfy the requirements of real-time process.Analysis based on both measurement data from Abilene and simulation experiments demonstrate that the method has very good detection and classification performance.

Keywords

• traffic anomaly
• online detection
• online classification
• incremental PCA
• incremental clustering