An Efficient Approach for Complex Data Summarization Using Multiview Clustering

Abstract

There is significant interest in the data mining and network management communities to efficiently analyse huge amount of network traffic, given the amount of network traffic generated even in small networks. Summarization is a primary data mining task for generating a concise yet informative summary of the given data and it is a research challenge to create summary from network traffic data. Existing summarization techniques are based on clustering and frequent itemset mining which lacks the ability to create summary for further data mining tasks such as anomaly detection. Additionally, for complex and high dimensional network traffic dataset, there is often no single clustering solution that explains the structure of the given data. In this paper, we investigate the use of multiview clustering to create meaningful summary from network traffic data in an efficient manner. We develop a mathematically sound approach to select the summary size using a sampling technique. The main contribution of this paper is to propose a summarization technique for use in anomaly detection. Additionally, we also propose a new metric to evaluate summary based on the presence of normal and anomalous data instances. We validate our proposed approach using the benchmark network traffic dataset.