A Unified Theory of Fuzzy c-Means Clustering Models with Improved Partition

摘要

This paper attempts to unify the theory of a certain class of modified variants and another class of manipulated versions of the fuzzy c-means algorithm. Starting from the objective function of the so-called fuzzy c-means algorithm with generalized improved partition (GIFP-FCM), and defining its rewarding term in a more flexible way, we obtain a unified algorithm that can model all algorithm variants in question including the wide family of suppressed and generalized suppressed FCM. Numerical tests were carried out to provide a comparison of the modeled algorithms in terms of accuracy and cluster size insensitivity. The suppression of the probabilistic fuzzy partition obtained at high values of the fuzzy exponent m proved the most effective.