A novel abnormal pattern identification method with ISODATA

초록

In this paper, we propose a new method of effectively identifying network traffic attacks by clustering their visualized forms according to the variance. The suggested method first visualizes the IP addresses and ports information of the transmitters and receivers into 2D images. Analyzing those images, it then extracts such features as linear patterns and high intensity values which indicates traffic attacks. Subsequently, it clusters the ports information by using an ISODATA algorithm and determines through an artificial neural network if the network traffic data contain DDoS, DoS, or Internet worm attacks. In experimental results, we show that our suggested ISODATA clustering-based algorithm can identify abnormal patterns of network traffic data successfully. © 2014 International Information Institute.

키워드