基于加权邻域依赖度的非单调性属性约简  

Non-monotonic Attribute Reduction Based on Weighted Neighborhood Dependency

作　　者：张敏 朱启兵[1] ZHANG Min;ZHU Qi-bing(College of Internet of Things,Jiangnan University,Wuxi Jiangsu 214122,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机仿真》2023年第3期395-399,503,共6页Computer Simulation

基　　金：国家自然科学基金(61772240)。

摘　　要：邻域粗糙集模型在对数据进行粒化时,只考虑了邻域半径内的样本个数,未考虑样本之间的分布信息,在一定程度上造成了原始数据信息的丢失。针对上述问题,定义加权邻域及加权邻域依赖度的概念,证明了加权邻域依赖度具有非单调性变化的特性。以加权邻域依赖度为度量基准,整合变精度和邻域粗糙集的优势,设计一种基于前向启发式的属性约简算法。通过在12个UCI数据集上的对比仿真,可以发现所提算法不仅可以有效提高分类精度,在一定程度上也降低了算法对变精度参数β的敏感性。When granulating data,neighborhood rough set model only considers the number of samples in the neighborhood radius,but does not consider the distribution information between samples,so it causes the loss of original data information to a certain extent.To solve this problem,the concepts of weighted neighborhood and weighted neighborhood dependency were defined in the paper,and it was proved that weighted neighborhood dependency has the property of non monotonicity.To solve this problem,a new attribute reduction algorithm based on forward heuristic was proposed,which integrated the advantages of variable precision and neighborhood rough set.Simulation results on 12 UCI datasets show that the proposed algorithm can not only effectively improve the classification accuracy,but also reduce the sensitivity of the algorithm to variable precision parameters to a certain extent.

关 键 词：邻域粗糙集 属性约简 距离 加权邻域 依赖度 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]