基于邻域关系粗糙集和不确定性的增量属性约简方法  

Incremental attribute reduction method based onneighborhood rough sets and uncertainty

作　　者：吴晓雪[1] 李艳[2] WU Xiaoxue;LI Yan(School of Mathematics and Information Science,Hebei University,Baoding 071002,China;Research Center for Applied Mathematics and Interdisciplinary Sciences,Beijing Normal University at Zhuhai,Zhuhai 519087,China)

机构地区：[1]河北大学数学与信息科学学院,河北保定071002 [2]北京师范大学(珠海)应用数学与交叉科学研究中心,广东珠海519000

出　　处：《西北大学学报（自然科学版）》2022年第5期753-764,共12页Journal of Northwest University（Natural Science Edition）

基　　金：国家自然科学基金(61976141);广东省自然科学基金(2018A0303130026);北京师范大学珠海分校教师科研能力促进计划。

摘　　要：当数据集发生变化时,基于粗糙集的增量方法可以对属性约简进行快速更新。考虑样本增加的动态情况,现有方法对添加的全部样本进行增量计算,时间消耗仍然较大。该文考虑样本的重要程度,认为不同区域的样本对约简更新的贡献程度不同,只选取贡献度大的样本参与约简的更新,从而有效降低计算量。在邻域粗糙集框架下,该文针对分类问题的连续值信息系统,根据样本的分布提出了基于不确定性和邻域关系粗糙集的增量属性约简方法。首先,利用不确定性和分类器结果对样本的贡献程度进行度量和类型划分;然后,针对不同类型的新增样本设计相应的处理策略,在此基础上提出新的增量属性约简算法;最后,在11个UCI数据集上进行大量实验,结果表明该方法与现有方法相比进一步降低了时间耗费,并保持了良好的分类精度和约简能力。Rough set-based incremental methods can perform fast updates to the attribute reduction when the dataset changes.Considering the dynamic situation of sample increase,the existing methods carry out incremental calculations on all the added samples,and the time consumption is still relatively large.In this paper,the importance of samples is taken into account,and we consider that samples in different regions contribute to the reduction update in different degrees.Only the added samples with large contribution are selected to participate in the reduction update,thereby effectively reducing the amount of computation.Under the framework of neighborhood rough sets,aiming at the continuous-valued information systems for classification problems,we propose an incremental attribute reduction method based on uncertainty and neighborhood rough sets according to the distribution of samples.Firstly,the uncertainty and classifier results are used to measure and classify the contribution of the added samples.Then,corresponding processing strategies are designed for different types of these samples,and a new incremental attribute reduction algorithm is proposed on this basis.Finally,extensive experiments are carried out on 11 UCI datasets,and the results show that this method further reduces the time consumption compared with the existing methods,and maintains good classification accuracy and reduction ability.

关 键 词：邻域关系粗糙集 属性约简 增量算法 不确定性 属性重要度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]