Computationally intensive parameter selection for clustering algorithms: The case of fuzzy c-means with tolerance

Parameter selection is a well-known problem in the fuzzy clustering community. In this paper, we propose to tackle this problem using a computationally intensive approach. We apply this approach to a new method for clustering recently introduced in the literature. It is the fuzzy c-means with tolerance. This method permits data to include some error, and this is modeled by moving data in a particular direction within a particular range when clusters are defined. The proper application of this approach needs the correct definition of the parameter κ. A value that might be different for each record and corresponds to the maximum shift allowed to the data. In this paper, we review this method and we study the definition of this parameter κ when the same value of κ is used for all data elements. Our approach is based on the analysis of sets of data with increasing noise and an exhaustive analysis of the behavior of the algorithm with different values of κ. The analysis is motivated in privacy preserving data mining. The same approach can be used for parameter selection in other clustering algorithms. © 2010 Wiley Periodicals, Inc.Partial support by the Spanish MEC (projects ARES – CONSOLIDER INGENIO 2010 CSD2007-00004 – and eAEGIS – TSI2007-65406-C03-02) is acknowledged.Peer Reviewe