基于因素空间理论的扫类连环多分类算法  

Algorithm of sweeping multiplex classification based on factor space theory

作　　者：曾繁慧[1,2] 王莹 汪培庄 孙慧 ZENG Fanhui;WANG Ying;WANG Peizhuang;SUN Hui(College of Science,Liaoning Technical University,Fuxin 123000,China;Institute of Intelligence Engineering and Mathematics,Liaoning Technical University,Fuxin 123000,China)

机构地区：[1]辽宁工程技术大学理学院,辽宁阜新123000 [2]辽宁工程技术大学智能工程与数学研究院,辽宁阜新123000

出　　处：《辽宁工程技术大学学报（自然科学版）》2024年第1期111-118,共8页Journal of Liaoning Technical University (Natural Science)

基　　金：辽宁省教育厅科学研究项目(LJ2019JL019);辽宁工程技术大学学科创新团队资助项目(LNTU20TD-35)。

摘　　要：为解决多分类问题,基于因素空间理论中因素显隐的思想,在扫类连环分类算法基础上,定义类别的合并,提出因素显隐的合并扫类连环分类方法,给出算法步骤,并用数值算例进行分析;定义类别的两两组合,提出因素显隐的两两扫类连环分类方法,给出算法步骤,并用数值算例进行分析。提出采用因素显隐的差额绝对值方法解决两个算法执行过程中出现的决策类别分不开的问题;对UCI数据集中3个实例与支持向量机作了算法对比分析,研究结果表明:提出的合并扫类连环分类方法、两两扫类连环分类方法实现了因素显隐,分类算法的精确度优于支持向量机。多分类学习的因素显隐研究结论拓展了因素空间的理论及应用研究。In order to solve the problem of multiple classification,based on the idea of explicit and implicit factor in factor space theory,the algorithm of sweeping class serial classification is proposed,the merging of categories is defined,and an explicit and implicit method of merging class sweep classification is proposed.The algorithm steps are given and analyzed by numerical examples.The pair wise combination is defined and the pairwise class sweep classification method with explicit and implicit factors is proposed,the algorithm steps are given and analyzed by numerical examples.During the execution of two algorithms,for the occurrence of decision categories can not be separated,the absolute value method of difference with explicit and implicit factors is proposed to solve the problem.Three examples in UCI dataset are compared with support vector machine,the results show that the proposed method of merging class sweep classification and the pairwise sweep classification method can realize factor explicit and implicit,and the accuracy of classification algorithm is better than that of the support vector machine.The conclusion of explicit and implicit study of factors in multi-classification learning expands the theoretical and applied research of factor space.

关 键 词：因素空间 因素显隐 扫类连环分类算法 合并扫类连环分类算法 两两扫类连环分类算法 差额绝对值法 

分 类 号：O29[理学—应用数学]