A semi-supervised clustering ensemble approach integrated constraint-based and metric-based

Abstract

Recently both semi-supervised clustering and clustering ensemble have received considerable attention due to their accurate and reliable performance. There are mainly two kinds of existing semi-supervised clustering algorithms called constraint-based and metric-based. In this paper, we present a semi-supervised clustering ensemble framework which takes both pairwise constraints and metric learning into account. Firstly, with the help of pairwise constraints or labeled data, there generates different base partitions by using constraint-based semi-supervised clustering algorithms and metric-based semi-supervised clustering algorithms respectively, in which the latter develops a new metric function for image data. Given the spatial particularity of image pixels, this metric measure considers the spatial distribution of surrounding pixels besides inherent features of pixels. And then the target clustering is obtained by integrating these base partitions into an ensemble function. Finally, we conduct the verification on standard data sets and image data sets. Both theoretical analysis and experimental results demonstrate that the proposed method produces considerable improvement in clustering accuracy with assistance of supervised information and yields superior clustering results over a number of representation clustering methods.