基于类内差和改进划分系数的聚类有效性函数

针对改进划分系数对模糊聚类有效性的判决并不十分理想,提出了将类内差和改进划分系数相结合的两个聚类有效性函数。该聚类有效性函数从数据聚类效果要求类内样本越相似而类间样本相差越大的观点出发,通过将反映数据聚类类内紧致性程度的类内差和类间分离性程度的改进划分系数相结合,并考虑到模糊C 均值聚类算法的适用条件作为构造聚类有效性函数的约束因子,得到新的聚类有效性标准。给出应用该函数进行模糊C 均值聚类有效性判决的具体步骤,通过仿真实验证明该有效性函数具有良好的分类性能。

Clustering validity function based on intra-variance and modified partition coefficient

Considering that modified partition coefficient used to make sure clustering validity is not fully ideal, two functions of clusering validity are proposed by combination of intra-variance with modified partition coefficient. In view of data clusering quality demanding intra-data with better comparability and inter-data with biggish separation, these clustering validity functions are formed by means of intra-variance which is a kind of geometry information to describe intra-data compactness combining with modified partition coefficient based on data maximum classification information, with considering application condition of fuzzy C-means clustering algorithm as restricted factor of clustering validity function. Detailed steps are given to apply these functions for fuzzy C-means clustering validity determinant. Experiment results show that the new methods have good classification performance.