基于异类粒球分离度的自适应属性约简  

Adaptive attribute reduction based on heterogeneous granular ball separability degree

作　　者：黄兵[1] 孙可 HUANG Bing;SUN Ke(School of Computer Science,Nanjing Audit University,Nanjing,Jiangsu 211815,China)

机构地区：[1]南京审计大学计算机学院

出　　处：《闽南师范大学学报（自然科学版）》2024年第3期1-16,共16页Journal of Minnan Normal University：Natural Science

基　　金：国家自然科学基金项目(62276136);江苏省研究生科研实践创新计划(KYCX23_2344)。

摘　　要：属性约简是处理大规模数据集的关键步骤,与传统的邻域粗糙集(NRS)相比,粒球邻域粗糙集(GBNRS)可以显著提高属性约简的性能.然而,目前GBNRS属性约简算法生成了太多不必要的粒球;从而极大降低了算法运行效率.文章首先定义了一种新的粒球质量指标来控制生成自适应数量的粒球;然后通过粒球对样本集进行划分,将不同类别的样本点放入不同类别的粒球;最后根据不同属性集合下粒球中正域样本的数量来进行前向属性约简.为了验证算法的有效性,在12个真实数据集上将提出的算法与其他NRS属性约简算法进行了对比实验.实验结果表明,所提出的算法有更高的精度和更快的运行效率.Attribute reduction is a key step in processing large-scale datasets.Compared with the traditional Neighborhood Rough Set(NRS),the Granular Ball Neighborhood Rough Set(GBNRS)can significantly improve the performance of attribute reduction.However,the current GBNRS attribute reduction algorithms generate too many unnecessary granular balls,which greatly reduces the efficiency of the algorithm.This paper first defines a new granular ball quality metric to control the generation of an adaptive number of granular balls.Then,it partitions the sample set using granular balls,placing sample points of different categories into corresponding granular balls.Finally,forward attribute reduction is performed based on the number of positive region samples within the granular balls under different attribute sets.To verify the effectiveness of the algorithm,comparative experiments were conducted with other NRS attribute reduction algorithms on 12 real datasets.The experimental results show that our proposed algorithm has higher accuracy and faster operational efficiency.

关 键 词：自适应粒球 属性约简 粒球邻域粗糙集 邻域粗糙集 分离度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]