邻域粗糙集的加权依赖度及其启发式约简算法

摘要

邻域粗糙集是数值型属性数据处理的有效工具.基于邻域粗糙集,传统依赖度及其约简未考虑邻域覆盖的绝对结构,由此文中建立加权依赖度及其启发式约简算法.首先,提出加权依赖度并得到其度量改进性与粒化单调性,定义相关的属性约简.然后,分析邻域半径的自适应取值,构造基于加权依赖度的启发式约简算法(NWDR).最后,在UCI数据集上进行对比实验,验证加权依赖度的单调性与NWDR的有效性.实验证明,加权依赖度改进传统依赖度的不确定性表示能力,NWDR具有较高的分类准确率与较强的应用适应性.%Neighborhood rough sets act as an effective tool for data processing of numeric attributes. According to neighborhood rough sets, the traditional dependency and its reduction rarely take the absolute structure of neighborhood covering into account. Therefore the weighted dependence and its heuristic reduction algorithm are established in this paper. Firstly,the weighted dependence is proposed to gain its measure improvement and granulation monotonicity, and its relevant attribute reduction is defined. Secondly, the self-adapting valuing of the neighborhood radius is analyzed, and the neighborhood weighted dependence reduction(NWDR algorithm) is constructed. Finally, contrast experiments on UCI datasets are implemented,and both the monotonicity of the weighted dependence and the effectiveness of NWDR are verified. The weighted dependence improves the uncertainty representation ability of the classical dependence, and the relevant NWDR exhibits higher classification accuracy and stronger application applicability.