基于样本对选择的分层分类特征选择算法研究  

Hierarchical Classification Feature Selection Algorithm Based on Sample Pair Selection

作　　者：折延宏[1] 武晋兰 贺晓丽[1] SHE Yanhong;WU Jinlan;HE Xiaoli(College of Science,Xi'an Shiyou University,Xi'an 710000,China;College of Computer,Xi'an Shiyou University,Xi’an 710000,China)

机构地区：[1]西安石油大学理学院,陕西西安710000 [2]西安石油大学计算机学院,陕西西安710000

出　　处：《昆明理工大学学报（自然科学版）》2022年第5期92-102,共11页Journal of Kunming University of Science and Technology(Natural Science)

基　　金：国家自然科学基金项目(61976244,12001422);陕西省自然科学基金项目(2021JQ-580);浙江省海洋大数据挖掘与应用重点实验室开放课题(OBDMA202105).

摘　　要：在大数据时代,数据的标签数量急剧增加,且标签集之间往往存在层次结构,利用层次结构进行大规模分类学习可有效解决维数灾难、类别不均衡问题,是近年来的研究热点.模糊粗糙集作为处理不确定性信息的有效工具,对于层次结构的描述有着特别的优势,本文给出了一种基于样本对选择的分层特征选择方法.通过将层次结构融入到目标概念的上、下近似之中,给出了一种新的模糊粗糙集模型,并研究了其详细性质,基于此,设计了一种基于样本对选择的特征选择算法,实验结果表明,本文所提出的算法在效率和准确性方面优于平面算法,从而为基于粒计算的分层特征选择提供了一种可能的框架.In the era of big data,the number of labels of data increases sharply,and there are often hierarchical structures between label sets.Using hierarchical structures to carry out large-scale classification learning can solve the problems of dimension disaster and category imbalance effectively,and has become a research concern in recent years.As an effective tool for processing uncertain information?fuzzy rough set has special advantages in describing hierarchical structure.In this paper,a method of hierarchical feature selection based on sample pair selection is proposed.A new kind of fuzzy rough set model is presented by integrating hierarchical structure into the target of the concept of upper and lower approximation,and its properties is studied in detail.Then a feature selection algorithm is designed based on sample pair selection.The experimental results show that the presented algorithm is better than that of plane algorithm in terms of efficiency and the accuracy,which provides a possible framework for hierarchical feature selection based on granular computing.

关 键 词：分层分类 模糊粗糙集 辨识矩阵 样本对选择 特征选择 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]