Robust Two-mode clustering

Starting from an extension of standard K-means for simultaneously clustering observations and features, namely Double K-Means (DKM) (Vichi, 2001), the model is developed in a probabilistic framework with a robustification necessary to take into account a certain amount of outlying observations assumed included in the data, that generally lead to unsatisfactory clustering results. An efficient algorithm is proposed and the advantages of using this approach are discussed. Key Words: Two-mode clustering, double k-means, disjoint principal component analysis, robustness. 1