Introducing interactive evolutionary computation in data clustering

Abstract

Data clustering consists in finding homogeneous groups in a dataset. The importance attributed to cluster analysis is related to its fundamental role in many knowledge fields. Often data clustering techniques are the ghost host of many innovative applications for a wide range of problems (i.e. biology, marketing, customers segmentation, intelligent machines, machine translation, etc.). Recently, there is an emerging interest in Data Clustering community to develop bio-inspired algorithms in order to find new methods for clustering. It is widely observed that bio-inspired algorithms and the Evolutionary Computation (EC) techniques reach solutions similar to others computational approaches but using a bigger computational power. This limitation represents a concrete obstacle to an extensive use of Evolutionary (or bio-inspired) approach to data clustering applications. In the present paper we propose to use Interactive Evolutionary Computation (IEC) techniques where a human being (the breeder) selects Cluster configurations (genotypes) on the basis of their graphical visualizations (phenotypes). We describe a first version of a software, called Revok, that implements the IEC basic principles applied to data clustering. In the conclusion section we outline the necessary steps to reach a mature IEC tool for data clustering.