Density based clustering with crowding differential evolution

The aim of this work is to analyze the applicability of crowding differential evolution to unsupervised clustering. The basic idea of this approach, interpreting the clustering problem as a multi-modal optimization one, is similar to that of unsupervised niche clustering proposed by Nasraoui et al.[10] but instead of evolving only the clusters centers and statistically estimating the other parameters (scales and orientation) we evolve both the centers and the param-eters of the clusters. Moreover, to simplify the evolutionary process, especially in the case of high-dimensional data, we evolve only hyper-ellipsoids parallel with the axes. In order to model rotated clusters we used a multi-center represen-tation, i.e. the cluster is covered by many normally oriented hyper-ellipsoids. Besides the fact that it simplifies the evolu-tionary process this multi-center representation allows de-scribing almost arbitrary shaped clusters. Preliminary ex-perimental results suggest that the proposed approach en-sures a reliable identification of clusters in noisy data pro-viding in the same time multi-center synthetic descriptions for them.