Soft ranking in clustering

Due to the diffusion of large-dimensional data sets (e.g., in DNA microarray or document organization and retrieval applications), there is a growing interest in clustering methods based on a proximity matrix. These have the advantage of being based on a data structure whose size only depends on cardinality, not dimensionality. In this paper we propose a clustering technique based on fuzzy ranks. The use of ranks helps to overcome several issues of large-dimensional data sets, whereas the fuzzy formulation is useful in encoding the information contained in the smallest entries of the proximity matrix. Comparative experiments are presented, using several standard hierarchical clustering techniques as a reference.