基于流形学习的自适应反馈聚类中心确定方法

An Adaptive Method for Determining the Feedback Clustering Center Based on Manifold Learning

K-means聚类是现行最为通用的聚类算法之一,但其在聚类过程中聚类中心不够稳定,针对这一问题,以流行聚类为依托,实现一种可以提升聚类中心稳定程度的反馈聚类方法。首先使用谱聚类对目标数据集初次聚类得到聚类集合,再根据目标集合最大子集数动态调整参数,多次迭代后获得聚类数集合表并对表中聚类值统计,最后计算表中加权平均值确定最终聚类数目。通过一百次迭代后的数据可知,使用加权平均值得到的聚类更稳定,反馈谱聚类迭代中位数和加权平均数的使用优化了流形聚类方法。

K-means clustering is one of the most the widely used clustering algorithms, but the clustering center is not stable enough in the clustering process. Based on the popular clustering, a feedback clustering method which could improve the stability of clustering center is achieved. Firstly, the spectral clustering is used to get the clustering set from the initial clustering to the target data set, and then dynamically adjust the parameters according to the maximum subsets number of the target set. After the iteration, clustering number set table is gotten and the clustering value in the table is calculated. At last, the weighted average in the tables is calculated to determine the number of fnal clusters. Experimental results of one hundred iterations show that the clustering method using the weighted average is more stable and the use of iterations median and weighted averages in feedback spectral clustering optimize the manifold clustering method.