A New Evolutionary Approach to Cluster Validation Index

Abstract

This paper presents a new evolutionary method for cluster validation index (CV!), which is a function of cluster-related parameters. The goal is to accurately discover the number of clusters of data sets. It is achieved by learning potential CVI from the training data sets using genetic programming (GP), and then estimating the optimal number of clusters by putting the parameters of test data sets into the learned CVI. In GP module, each chromosome encodes a possible CVI as a function of the number of clusters, density measure of clusters, and some random factors. The fitness function evaluating the quality of candidates is defined by the difference between the actual number of clusters of each training data set and the number of clusters computed by each chromosome. Due to the adaptive nature of GP, the proposed evolutionary CVI is reliable and robust in various types of data sets. Experimental results provide grounds for the dominance of our method over several widely-known CVIs.