基于矩阵方法的区分度增量式属性约简算法  

A DISCRIMINATION DEGREE INCREMENTAL ATTRIBUTE REDUCTION BASED ON MATRIX METHOD

作　　者：沈玉峰[1] Shen Yufeng(College of Computer Engineering,Anhui Sanlian University,Hefei 230601,Anhui,China)

机构地区：[1]安徽三联学院计算机工程学院,安徽合肥230601

出　　处：《计算机应用与软件》2020年第9期235-245,257,共12页Computer Applications and Software

基　　金：安徽省重点科研项目(KJ2019A0903)。

摘　　要：属性约简是粗糙集理论在机器学习领域的重要应用。通过对传统的属性约简算法构造增量式学习,从而提高了动态数据环境下的属性约简性能。区分度作为一种重要的属性集评估函数,目前已成功地运用于属性约简的构造。在基于区分度属性约简的基础上,提出一种区分度的增量式属性约简算法。通过矩阵的方法去表示区分度度量,在矩阵表示的基础上,进一步研究信息系统对象增加和减少时区分度的增量式学习,根据这种增量式学习提出对应的增量式属性约简算法。UCI数据集的实验结果表明,所提出的增量式属性约简算法比非增量式算法具有更高的动态数据属性约简性能。Attribute reduction is an important application of rough set theory in machine learning.By constructing incremental learning for traditional attribute reduction algorithm,the performance of attribute reduction in dynamic data environment is improved.As an important attribute set evaluation function,discrimination degree has been successfully applied to attribute reduction.This paper proposes an incremental attribute reduction algorithm based on discriminant degree.The matrix method was used to represent the discrimination degree.Then,on the basis of the matrix representation,I further studied the incremental learning of the discrimination degree when the object of information system increased or decreased.Through the incremental learning,the corresponding incremental attribute reduction algorithm was proposed.The experimental results on UCI datasets show that my algorithm has better performance than the non-incremental algorithm in dynamic data attribute reduction.

关 键 词：属性约简 增量式学习 信息系统 动态变化 矩阵 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]