SMS – propozycja nowego algorytmu analizy skupień
SMS – proposal of new clustering algorithm

Summary/Abstract: Spectral clustering [Ng, Jordan, Weiss 2002; von Luxburg 2006] and mean shift clustering {Wang, Xiu, Damar 2007] are two relatively new approaches in cluster analysis, giving, especially for clusters of unusual shapes, better results than classical methods such as k-means, k-medoids or hierarchical agglomerative methods. The article contains a proposal for algorithm with the working name − SMS (Spectral-Mean Shift) that combines features of both approaches, distinguishing among other cluster analysis algorithms with: – the possibility of cluster analysis of unusual shapes, – the ability to automatically identify the number of clusters, – better resistance to interference (noisy) variables.

• Details
• Contents