一种改进的K-均值聚类算法的研究

On the Improvement of K-means Clustering Algorithm

聚类分析在科研和商业应用中都有着非常重要的作用。K-均值聚类算法是一种基于样本间相似性度量的间接聚类方法,其不足之处是,它采用均值作为一类的代表点,一个点往往不能充分反映该类的模式分布结构,从而损失了很多有用的信息。研究了一种改进的K-均值聚类算法,在求样本间距离时,采用核函数距离代替欧氏距离,考虑了各模式间的相关性。试验结果表明,利用改进的K-均值聚类算法,聚类结果的准确率更高,更稳定。

Clustering analysis plays an important role in scientific research and commercial application.The K-means algorithm is the indirect clustering algorithm based upon comparability measurement between points,but it has some faults,it uses means as representative point,the point can’t figure the distributional structure of the mode,thus some important information is missed.An improved K-means clustering algorithm is studied.Kernel function distance is used to replace Euclidean distance.Experimental results show that clustering effect is higher veracity and more steady with improvement of K-means clustering algorithm.