ABSTRACT
Fuzzy clustering algorithms considered in Chapter 9 may be generalized in various aspects. One possibility is to consider a slight modification of the criterion function. In this way an infinite family of GFNM objective functions is obtained. The introduced weighting exponent parameter controls the fuzziness degree of the resulting partition. The limit properties of the GFNM family are considered. Other GFNM variations concern the prototypes and distances. In this chapter, a fuzzy clustering procedure using sets as class prototypes is considered. This representation is suitable when no information concerning the geometric shape of the clusters is available or the point prototype representation is too coarse. A clustering procedure using Lp metric is considered as an alternative to the clustering with Euclidean metric.
