摘要
传统邻域粗糙集需指定半径或通过搜索方式找出适用于问题求解的半径,这在数据预处理过程中会带来极大的时间消耗。而粒球粗糙集方法则能够依据数据分布,自适应地生成合适的粒结构。以粒球的纯度为度量准则,粒球粗糙集方法亦为属性约简问题的研究引入新的思路。利用前向贪心搜索求解约简时,需尝试计算每一个候选属性被加入约简池后所引起的粒球纯度的变化,这为算法的执行效率带来了严峻挑战。为解决这一问题,在前向贪心搜索进程中提出了属性划分策略,其本质是将所有属性划分成不同的组,从而能够压缩候选属性的搜索空间,以达到快速求解约简的目的。使用了10组UCI数据集,最终的实验结果说明,相较于传统邻域粗糙集约简以及基于纯度的粒球粗糙集约简,引入属性划分策略后,能够极大地提升粒球粗糙集约简求解的时间效率。
It is frequently required to specify or search the appropriate radii when the traditional neighborhood rough set is used.It is of very time consuming for data preprocessing.However,the granular ball rough set can adaptively generate granular structure according to the distribution of the data itself.If the purity of granular ball is regarded as the criterion,then granular ball rough set will provide new ideas for the research of attribute reduction.By using the forward greedy searching for deriving reduct,it is necessary to calculate the variation of the purity if each candidate attribute is added into the pool set,which will pose serious challenges for the efficiency of the algorithm.To solve this problem,an attribute partition strategy is introduced into the forward greedy searching.The essence of such strategy can divide the raw attributes into different groupings and the searching space of candidate attributes can be compressed for the purpose of quickly obtaining reduct.The experimental results from the 10 UCI data sets demonstrate that by comparing with the attribute reduction approaches based on traditional neighborhood rough set and granular ball rough set,the attribute partition based approach can significantly improve the time efficiency of calculating reduct in terms of granular ball rough set.
作者
巴婧
陈妍
杨习贝
Ba Jing;Chen Yan;Yang Xibei(School of Computer,Jiangsu University of Science and Technology,Zhenjiang 212100,China)
出处
《南京理工大学学报》
CAS
CSCD
北大核心
2021年第4期394-400,共7页
Journal of Nanjing University of Science and Technology
基金
国家自然科学基金(62076111,62006099,62006128,61906078)。
关键词
属性划分
属性约简
粒球
粗糙集
attribute partition
attribute reduction
granular ball
rough sets
