Block fuzzy k-modes clustering algorithm

Most clustering algorithms, such as k-means and fuzzy c-means (FCM), are used to cluster a set of objects based on a function of dissimilarities between objects. However, clustering on attribute variables of objects may give more cluster information. Thus, to have a clustering algorithm that can be designated to construct simultaneously an optimal partition of objects and also attribute variables into homogeneous block is important. This kind of clustering was called block clustering (see Duffy and Quiroz, 1991). Recently, Govaert and Nadif (2003) proposed a block classification EM (block CEM) algorithm and then proposed block fuzzy c-methods (block FCM) in 2006. In this paper, based on Huang and Ng´s (1999) fuzzy k-modes (FKM) method, we propose a block FKM clustering algorithm. Several examples are used to make the comparisons between block FCM and the proposed block FKM.