The hyperellipsoidal clustering using genetic algorithm

Hyperellipsoidal clustering can characterize the distribution of the clusters better than common hyperspherical clustering. In this paper, it is proved that the direct application of Mahalanobis distance instead of Euclidean distance as the similarity measure cannot acquire the hyperellipsoidal clustering. Based on the analysis a new similarity measure suitable to hyperellipsoidal clustering is presented and a genetic algorithm is applied to optimize the modified clustering cost function. The simulation experiments show the efficiency of the new algorithm