Improvement of the fuzzy C-Means clustering algorithm with adaptive learning of the dissimilarities among categorical feature values

In [1], recently we proposed a generalization of the frequency-based cluster prototype [2-4], in the same framework of the Fuzzy C-Means clustering algorithm, for the objects of mixed features. In the generalization, a general dissimilarity measure, not the simple matching dissimilarity, is assumed for each categorical feature. In this paper we develop an adaptive method to learn dissimilarity measures for categorical features. We include the method into the framework of the Fuzzy C-Means algorithm so that the clustering algorithm can use the dissimilarity measures rather than the simple matching dissimilarity measure for categorical features. Through the experiments over real object sets, we show the clustering quality becomes better.