Fuzzy Co-clustering Algorithm for Multi-source Data

Abstract

The development of information and com- munication technology has motivated multi- source data to become more common and publicly available. Compared to traditional data that describe objects from a single- source, multi-source data is semantically richer, more useful, however many-feature, more uncertain, and complex. Since tra- ditional clustering algorithms cannot han- dle such data, multi-source clustering has become a research hotspot. Most existing multi-source clustering methods are devel- oped from single-source clustering by ex- tending the objective function or building combination models. In fact, the fuzzy clus- tering methods handle the uncertainty data better than the hard clustering methods. Re- cently, fuzzy co-clustering has proven effec- tive in the many-feature data processing due to the possibility of isolating the uncertainty present in each feature. In this paper, a novel multi-source data mining algorithm based on a modified fuzzy co-clustering algorithm and two penalty terms is proposed, which is called Multi-source Fuzzy Co-clustering Algorithm (MSFCOC)Experimental results on various multi-source datasets indicate that the proposed MSFCOC algorithm outper- forms existing state-of-the-art clustering al- gorithms.

Keywords: Data mining, multi-source, fuzzy co-clustering, multi-view, multi- subspace.