Visual decision support for ensemble clustering

Abstract

The continuing growth of data leads to major challenges for data clustering in scientific data management. Clustering algorithms must handle high data volumes/dimensionality, while users need assistance during their analyses. Ensemble clustering provides robust, high-quality results and eases the algorithm selection and parameterization. Drawbacks of available concepts are the lack of facilities for result adjustment and the missing support for result interpretation. To tackle these issues, we have already published an extended algorithm for ensemble clustering that uses soft clusterings. In this paper, we propose a novel visualization, tightly coupled to this algorithm, that provides assistance for result adjustments and allows the interpretation of clusterings for data sets of arbitrary size. © 2010 Springer-Verlag Berlin Heidelberg.