A genetic approach to fuzzy clustering with a validity measure fitness function

This paper presents an extension to the genetic fuzzy clustering algorithm proposed by the authors. The original algorithm, which combines the powerful search technique of genetic algorithms with the fuzzy c-means (FCM) algorithm, is extended such that the FCM algorithm was totally embedded in the genetic operators design. Two objective functions are applied as fitness functions: the performance index of a P fuzzy c-partition J(m)(P), used on the FCM algorithm, and the partition coefficient F-c(P), a function commonly used as a measure of cluster validity. The fuzzy c-means and the new proposal for the genetic fuzzy clustering algorithm were compared on generating multiple prototypes. The experimental results show that the use of genetic search improves the quality of the clustering solutions and that the partition coefficient F-c(P) is a better measure for clustering than the performance index J(m)(P).