给出了一种新的无监督聚类算法,但这种算法并非是基于目标函数的聚类算法,而是对数据直接设计一种迭代运算,以使数据在保持类特征的情况下进行重新组合最终达到分类的目的.通过对一类数据的实验表明,该算法在无监督给出类数方面具有较好的鲁棒性;另外,该算法在数据的准确归类、无监督聚类、确定性,以及对特殊类分布的适用性等方面均优于HCM和FCM算法.

In this paper, a new algorithm for unsupervised Clustering analysis is proposed, through a new kind of iterative activation the examples of a cluster are moved inside to the center of the gravity of the cluster together. Through this method correct number of clusters could be got. Because each sample moves only in its own cluster while iterating, we can correctly tell which cluster a sample should belong to. The experiments show that the new algorithm has better results in several aspects than HCM and FCM algorithms, such as unsupervised clustering, correct clustering, clustering capability for special data which HCM and FCM algorithms can not cluster. The new algorithm is an unsupervised clustering algorithm but HCM and FCM algorithms need correct number of clusters before iterative activation.