可能性划分系数和模糊变差相结合的聚类有效性函数

CLUSTERING VALIDITY FUNCTION BASED ON POSSIBILISTIC PARTITION COEFFICIENT COMBINED WITH FUZZY VARIATION

基于可能性分布描述因子定义的可能性划分系数有随类数增加而单调递减的趋势,缺乏与数据集几何结构的直接联系。该文考虑到数据集的几何结构信息,对可能性划分系数进行改进,提出了新的聚类有效性标准。实验结果表明,该文提出的方法具有良好的分类性能。

Possibilistic partition coefficient which based on possibilistic distribution descriptor Ihas decreasing tendency as the classification number increasing and does not directly relate to |the geometry structure of data set. To consider the geometry structure information of data fset, new clustering validity functions are defined by modify possibilistic partition coefficient. 5Experimental results show that the new methods have good classification performance. I