关键词: K-means算法, 方差, 加权, 最大最小法, 簇初始中心点

Abstract: The selection of initial cluster centroids in the classical K-means algorithm is random, which causes that the clustering results vary with different selections of cluster centroids. Thereby many selection approaches of initial centroids are devised and applied. However, most of them are affected by parameters design and parameter values. To overcome this problem, this paper proposes a novel initial cluster centroids selection algorithm, called WLV-K-means (weighted local variance K-means). The WLV-K-means algorithm employs the weighted local variance to measure the density of each sample, which can find samples with higher density. This algorithm also uses the improved max-min method to select cluster centroid heuristically. The experiments are made on UCI datasets and the results show that the WLV-K-means algorithm outperforms some improved K-means algorithms and is more stable and robust.

Key words: K-means algorithm, variance, weighting, max-min method, initial cluster centroids