SE-Stream: Dimension projection for evolution-based clustering of high dimensional data streams

Abstract

Evaluation-based stream clustering method supports the monitoring and detection of the change in clustering structure. E-Stream is an evolution-based stream clustering method that supports different types of clustering structure evolution which are appearance, disappearance, self-evolution, merge and split. However, its runtime increases and its performance drops when face with high dimensional data. High dimensional data leads to more complexity in the clustering methods. In this paper, we present SE-Stream which extends E-Stream in order to support high dimensional data streams. A projected clustering technique to determine specific subset of dimensions for each cluster is proposed. The proposed technique reduces complexity of calculation. Each cluster describes itself by a set of selected dimensions. Experimental results show that SE-Stream gives better cluster quality compared with E-Stream and HP-Stream, a state of the art algorithm for projected clustering of high dimensional data streams. Further, it gives better execution time compared with E-Stream and comparable execution time compared with HP-Stream.