Efficient Unsupervised Clustering through Intelligent Optimization

G.S. Antzoulatos, E.K. Ikonomakis, and M.N. Vrahatis (Greece)


Computational Intelligence, Evolutionary Computation, Data Clustering, Differential Evolution, Window Density Function


A novel methodology for unsupervised data clustering based on Evolutionary Computation, named “Intelligent Unsupervised Clustering” (IUC) is introduced. IUC searches for the “optimal clusters’ representatives” using Evolutionary Algorithms (EAs) and utilising a Window Density Function (WDF) as an objective function. EAs ensure that the representative is posed in a region of points of high density. IUC aims in finding a highly dense hyper rectangle around the cluster’s representative, that captures a part of cluster. Therefore, IUC uses a windowing technique and gradually enlarges a window, which is centered on the best individual generated from the EA. This process continues until the increase of the value of WDF does not change “drastically”. The whole process is repeated on the unclustered data, until all the clusters are discovered. The quality of clustering, delivered by the IUC, is compared with well-known clustering algorithms and the experimental results illustrate its efficiency and accuracy.

Important Links:

Go Back