A Framework for Monitoring Clustering Stability Over Time

Abstract

Mining data streams and arriving at intelligent decisions is becoming more and more important nowadays as a lot of applications produce large volume data streams. Data stream clustering has been considered to be very useful for online analysis of streams. Monitoring the cluster transitions over time provide good insight into the evolving nature of the data stream. This paper introduces a framework for monitoring the stability of individual clusters and clusterings over time, along with the progress of the stream. Tracking the historical evolution of clustering structures is the main focus of this framework. Two real-world datasets have been used for conducting the experiments. The results point up the fact that monitoring the stability of clustering structures will help to get an important hint of the physical events happening in the environment. This information can be used to predict the future clustering structure changes and in turn the upcoming events.