一种新的选取初始聚类中心的K-means算法

A New K-means Algorithm for Selecting Initial Clustering Center

文章在2014年提出的K-means初始聚类中心选取算法的基础上进行改进。通过计算样本间的相异度函数,求出每个样本的相异度参数,选取最大相异度参数值所对应的样本作为初始聚类中心。当最大相异度参数不唯一时,提出了一种合理选取最大相异度参数值的解决方案,依次求出K个初始聚类中心,由此提出了一种选取初始聚类中心的改进算法。实验证明,所提出的改进算法与原算法相比具有更高的准确率,并且明显减少了迭代次数。

This paper improves the algorithm based on the K-means initial clustering center selection algorithm proposed in 2014. By calculating the dissimilarity function between samples, the parameter of dissimilarity of each sample is obtained, and the sample corresponding to the maximum dissimilarity parameter is selected as the initial clustering center. The paper also puts forward a solution of choosing the parameter value of maximum dissimilarity reasonably when the maximum dissimilarity parameter is not unique, and then determines the K initial clustering centers successively, on the basis of which an improved algorithm is presented for selecting the initial clustering center.