特定类的代价敏感近似属性约简  

Cost sensitive approximate attribute reduction for specific classes

作　　者：胡军[1,2] 黄小涵 HU Jun;HUANG Xiaohan(Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,China;School of Computer Science and Technology,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学计算智能重庆市重点实验室,重庆400065 [2]重庆邮电大学计算机科学与技术学院,重庆400065

出　　处：《智能系统学报》2024年第6期1468-1478,共11页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金项目(62221005,62276038);重庆市自然科学基金项目(cstc2021ycjh-bgzxm0013);重庆市教委重点合作项目(HZ2021008).

摘　　要：特定类属性约简指针对特定决策类提供对应约简集的属性约简,现有特定类属性约简方法过于严苛,限制其在一些场景下的应用。针对存在噪声的数据,提出一种特定类的代价敏感近似属性约简方法。该方法首先结合正域与边界域信息定义特定类的相对不确定度,然后利用相对不确定度与测试代价计算属性重要度,进而根据属性重要度选择属性,并通过放松相对不确定度来避免冗余属性的加入,最后给出了特定类的代价敏感近似启发式属性约简算法。实验结果表明,所提方法与同类方法相比能够在保持甚至提升约简质量的同时获得更精简的约简集,并且约简集的测试代价相对更小。Class-specific attribute reduction refers to reducing attributes that are provided specifically for a given decision class.Existing class-specific attribute reduction methods are often too strict,which limits their applicability in certain scenarios.For noisy data,this paper proposes a cost-sensitive approximate attribute reduction method tailored for specific classes.First,the method combines information from the positive and boundary regions to define the relative uncertainty for a specific class.Then,attribute importance is calculated using relative uncertainty and test cost,allowing for attribute selection based on importance and avoiding the inclusion of redundant attributes by relaxing the relative uncertainty.Finally,the study introduces a cost-sensitive approximate heuristic attribute reduction for specific classes.Experimental results show that the proposed method can maintain or even improve the reduction quality while achieving a more streamlined reduction compared to other methods,with a relatively lower test cost for the reduction set.

关 键 词：粗糙集 不确定信息 特定类 相对不确定度 属性重要度 测试代价敏感 近似属性约简 启发式算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]