变权划分系数及其分类效果评价

针对模糊C-均值聚类算法对初始化分类参数(包括起始聚类中心位置和初始化分类隶属度矩阵)的选择比较敏感而导致分类结果差异性较大，以及错误分类会给解决实际问题带来难以预料后果的不足，本文从反映数据聚类后类间分离性测度的划分系数入手，提出了可变加权划分系数的新概念，并用于数据分类效果的评价。实验结果表明，本文提出的评价方法不仅是可行的，而且比模糊C-均值聚类算法的目标函数作为数据分类效果的评价准则更好。

Changeable Weighting Partition Coefficient and Its Classifying Quality Evaluation

Fuzzy C means algorithm for the initialized classification parameter (including initialized clustering centers and fuzzy membership function matrix) is sensitivity to data classifying quality, and different initialized classification parameters generate a classifying result with bigger otherness. This paper puts forward a changeable weighting partition coefficient based on partition coefficient reflecting inter class separation, and a new evaluating criterion based on changeable weighting partition coefficient is proposed to assess the data classifying quality. Experimental results show that the evaluating criterion is feasible, and it is better than objective function in fuzzy C means algorithm to evaluate data classifying quality.