Fuzzy Double Clustering: A Robust Proposal

In this paper a robust fuzzy methodology for simultaneously clustering objects and variables is proposed. Starting from Double k-Means, different fuzzy generalizations for categorical multivariate data have been proposed in literature which are not appropriate for heterogeneous two-mode datasets, especially if outliers occur. In practice, in these cases, the existing fuzzy procedures do not recognize them. In order to overcome that inconvenience and to take into account a certain amount of outlying observations a new fuzzy approach with noise clusters for the objects and variables is introduced and discussed.