基于Markov Blanket的模糊差别矩阵属性约简算法  

Markov Blanket-based Attribute Reduction Algorithm with Fuzzy Discernibility Matrix

作　　者：胡蝶怡 周杰 高灿[1,2] HU Die-yi;ZHOU Jie;GAO Can(College of Computer Science and Software Engineering,Shenzhen University,Shenzhen 518060,China;Key Laboratory of Intelligent Information Processing(Shenzhen University),Guangdong Province,Shenzhen 518060,China)

机构地区：[1]深圳大学计算机与软件学院,广东深圳518060 [2]广东省智能信息处理重点实验室(深圳大学),广东深圳518060

出　　处：《模糊系统与数学》2022年第6期54-63,共10页Fuzzy Systems and Mathematics

基　　金：国家自然科学基金资助项目(61806127,62076164)。

摘　　要：模糊差别矩阵属性约简是一种广泛使用的模糊粗糙集属性约简方法。然而已有方法大多采用启发式贪婪策略,属性约简率低且约简质量差。本文结合Markov Blanket概念,提出基于模糊差别矩阵的属性约简算法。首先,为了避免约简选择过多属性的问题,提出了利用Markov Blanket性质的迭代后向删除属性约简算法,可以有效删除低频率的相对冗余属性。其次,提出了一种基于Markov Blanket的双向搜索启发式属性约简算法,通过迭代前向添加高频属性和后向删除低频率策略来提升属性约简计算效率。在UCI数据集上实验表明,相比于其他模糊差别矩阵算法,所提出的算法能得到更优的约简结果。Fuzzy discernibility matrix attribute reduction is a widely used attribute reduction method of fuzzy rough sets. However, most of the existing methods use heuristic greedy strategy, which has low attribute reduction rate and poor reduction quality. Combined with the concept of Markov Blanket, this paper proposes an attribute reduction algorithm based on Fuzzy discernibility matrix. Firstly, in order to avoid the problem of selecting too many attributes, an iterative backward deletion attribute reduction algorithm based on Markov Blanket is proposed to effectively delete low-frequency relatively redundant attributes. Secondly, a bidirectional search heuristic attribute reduction algorithm based on Markov Blanket is proposed, which improves the computational efficiency of attribute reduction by adding high-frequency attributes forward and deleting low-frequency attributes backward.Experiments on UCI data sets show that the proposed algorithm can get better reduction results than other fuzzy discernibility matrix algorithms.

关 键 词：模糊粗糙集 属性约简 模糊差别矩阵 Markov Blanket 双向搜索 

分 类 号：O159[理学—数学]