混合属性对象的类别关系修正的模糊聚类方法

Category relations modification fuzzy clustering of mixed data

针对模糊聚类分析在处理混合条件属性数据时存在的不足,提出一种基于类别关系修正的集成方法。首先对分类条件属性特征参数采用熵表示类别隶属度,数值条件属性特征参数采用欧氏距离结合熵表示相似性;然后定义数据的混合类别模糊度及具体单个类别的模糊可信度,并由两者数值共同生成类别修正的线性、指数及对数变化的三种关系;最终通过类别关系修正值来衡量数据对象的类别模糊度。与多种已有的聚类集成方法对比实验表明,该方法具有优良的聚类性能。

In order to solve the weakness of fuzzy clustering analysis for mixed data set, this paper proposed a method of en- semble based on category relation correction. Firstly, the classification attribution of characteristic parameter used entropy to ex- press. The similarity of numerical characteristic parameter used Euclidean distance and entropy to express. Secondly, this paper defined the mixed category fuzzy and the single category fuzzy confidence, and they made the category relations modification for linear,exponential and logarithmic changes. Lastly, this paper used category relations modification to measure the category fuzzy. Experimental results show that the method proposed is better than the current methods for clustering ensemble.