Farthest Centroids Divisive Clustering

A method is presented to partition a given set of data entries embedded in Euclidean space by recursively bisecting clusters into smaller ones. The initial set is subdivided into two subsets whose centroids are farthest from each other, and the process is repeated recursively on each subset. The bisection task can be formulated as an integer programming problem, which is NP-hard. Instead, an approximate algorithm based on a spectral approach is given. Experimental evidence shows that the clustering method often outperforms a standard spectral clustering method, but at a higher computational cost. The paper also discusses how to improve the standard K-means algorithm, a successful clustering method that is sensitive to initialization. It is shown that the quality of clustering resulting from the K-means technique can be enhanced by using the proposed algorithm for its initialization.

• Type: Conference
• Year: 2008
• Where: ICMLA
• Authors: Haw-ren Fang, Yousef Saad
• Keywords: Clustering Method, ICMLA 2008, Machine Learning, Spectral Clustering Method, Successful Clustering Method