Finding the Optimal Clusters from Discrete Data using Evolutionary Clustering Algorithm

Abstract

To find the natural number of clusters in a dataset has been a difficult issue in many clustering algorithms. It is complex to determine the desired number of clusters. Even if a hierarchical tree of clusters is given, it is still hard to determine the cut points. The proposed algorithm is modified to find the optimal number of clusters. It is varied in several ways. It allows a variable number of clusters in the chromosomes. It introduces split and merge mutation and new crossover operations. In addition, the fitness function is enhanced to give a fair evaluation. The experimental results illustrate the proposed algorithm can successfully find the correct number of clusters in many datasets.

Keywords: Clustering algorithm, Fitness Function