新的模糊聚类有效性指标

Novel Validity Index for Fuzzy Clustering

为了更有效地确定模糊聚类算法的最佳聚类数,提出了一个新的有效性指标.该指标考虑了聚类的紧致度、重叠度、分离度.紧致度是用来衡量类内相似程度;重叠度是用来衡量类间的重叠程度;分离度度用来衡量类间的分离程度.利用该指标可以找到最符合数据自然分布的聚类数.实验结果证明,新的指标均能发现最优聚类数,从而克服了模糊c均值(FCM)算法聚类数需要预先设定的缺点,并能够准确地判断含有交叠子类的最佳聚类数.

A novel validity indice is proposed to determine the optimal number of clusters for fuzzy clustering. The novel validity indice considers the degree of compactness, the degree of overlapping, and the degree of separation . The compactness measures the similarity within a cluster. The degree of overlapping measures the overlap between clusters. Meanwhile, the degree of separation is used to measure the degree of the clear between clusters. The optimal cluster number can be effectively found by the proposed index. The experimental results show that the optimal cluster number are obtained which are general used in FCM algorithm, the new index overcomes the shortcomings of FCM that the cluster number must be pre-assigned and works well in the situations when there are overlapping subcluster in the clusters.