网格质心运动的聚类初始化方法

Initializing cluster method with motion of grid's center

针对以k-means为代表的分割聚类算法初始参数的很难选取这一难题,提出基于网格质心运动的初始化算法。划分网格后,定义网格的质量,利用物质质心运动理论,提取样本的聚类中心,并由此确定样本分类数k。实验表明,该算法可以有效地提取初始聚类中心,消除噪声点,可以提高后续聚类分析的效果和效率。

In order to solve the problems that how to set the number of classifications and the initial cluster center for k-means algorithm that is regarded as exemplifying of partitioning algorithms,an initial algorithm based on center motion is proposed.After gridding and defining mass of grid,motion theorem of the mass center is used to extract the centers of clustering samples,and then classifications are determined.Experiments on synthetic datasets show that compared with current approaches,this method can extract the centers of clustering samples more validly,restrain noise,improve clustering effect,and have good efficiency for cluster analysis.