基于邻域区分度的不完备混合数据属性约简方法

Attribute reduction method for incomplete mixed data based on neighborhood discernibility degree

针对不完备混合数据,提出一种基于邻域区分度的属性约简方法.首先,在不完备混合邻域决策系统中,针对3种数据类型定义新的距离函数,由此给出邻域、邻域容差关系及其邻域上下近似集等概念,构造邻域粗糙集模型.然后,在不完备混合邻域决策系统中定义区分关系,给出邻域区分度、相对邻域区分度等度量,并讨论相关性质及定理.最后,基于相对邻域区分度定义不完备混合邻域决策系统的属性约简集、属性重要度等概念,设计一种启发式的不完备混合数据属性约简算法.8个公共数据集上的实验结果表明,所提出的属性约简算法可以获取最优/次优属性子集,且具有较好的分类精度.

An attribute reduction method based on neighborhood discrimination is proposed for incomplete mixed data.First,new distance functions are defined for three types of data in incomplete mixed neighborhood decision systems,the concepts of neighborhood,neighborhood tolerance relation and neighborhood upper and lower approximate sets are given,and then the neighborhood rough set model is constructed.Second,the discrimination relation is defined in incomplete mixed neighborhood decision systems,some uncertainty measures such as neighborhood discrimination degree,relative neighborhood discrimination degree and so on are developed,and then the corresponding properties and theorems are discussed.Finally,based on the relative neighborhood distinguish degree,the attribute reduction set and attribute significance are defined in incomplete mixed neighborhood decision systems,and then a heuristic attribute reduction algorithm for incomplete mixed data is designed.The experimental results on eight public datasets show that the proposed attribute reduction algorithm can obtain optimal/suboptimal attribute subset,and exhibit better classification accuracy.