Clustering stability evaluation method based on SSIM

Abstract

In the clustering validity analysis, three main methods including intra-class cohesion, inter-class separation, and artificial judgment index can be used to evaluate the clustering results. If the clustering result is efficient, it means that the clustering stability is better. However, when those methods are used, it is essential to provide the sample data or clustering algorithms in advance. This paper proposes a clustering stability evaluation method based on the Elliptic Fourier Descriptor structural similarity index (EFD-SSIM), which can evaluate the clustering stability only when the clustering result is available. Its mechanism is that cluster is mapped into 2D graphics, and the degree of intra-class cohesion is measured based on the structural similarity (SSIM) on the graphics. As shown by the experimental results, EFD-SSIM has a good evaluation effect and it is consistent with the existing effectiveness evaluation indices of the clustering algorithm.