APP下载

基于集成学习的三支决策模型

2025-08-27王迪钱进郑明晨

郑州大学学报(理学版) 2025年6期

中图分类号:TP18 文献标志码:A DOI:10.13705/j. issn.1671-6841.2024095

文章编号:1671-6841(2025)06-0042-09

Abstract: Three-way decision model is an effective way to deal with complex decision problems by categorizing objects into three distinct decision regions.However,the existing three-way decision models often rely on a single decision criterion,limiting their efectiveness in handling complicated decision problems.To enhance the robustnessand accuracy of the decision-making process,a novel three-way decision model based on ensemble learning was proposed.Firstly,diffrent decision criteria were adopted in the decision-making process to obtain diferent three-way decision results. Then,inspired by the idea of pessimistic multi-granular rough sets,the consensus sets of the three decision regions were acquired by using basic operations between sets,respectively. Next,the k -means algorithm was utilized to divide the objects in the inconsistent set into three disjoint subsets according to their similarities. These subsets were then added to their respective consensus sets to obtain the final three-way decision results. The effcacy of this newly proposed model were substantiated through extensive experiments on diffrent datasets.Based on experimental results across various datasets,the newly proposed model achieved higher classification accuracy and comprehensive evaluation index.Additionally,the new three-way decision model occupied a smaller boundary region compared with other traditional three-way decision models.

Key words: three-way decision; ensemble learning; cluster ensemble; cluster analysis

0 引言

随着数据规模的不断增加,当今决策者面临着更加复杂的决策环境。决策者如何快速做出正确决策是一项值得研究的问题。Yao[为了降低决策过程风险,提出了三支决策模型,其基本思想是将整体划分为三个互不相交的子区域,对每个区域采取不同的决策行为。三支决策对决策问题的求解方法与人类的认知过程相符合,因此常用于复杂问题的分析和求解[2-4] 。

自三支决策模型提出以来,众多研究者已经对其进行了不断完善和发展。Qian等5考虑将多视角思想与三支决策模型相结合,提出了多粒度序贯三支决策,并给出了五种不同的粒度聚合策略。Jia等[将三支决策与多属性决策相结合,利用多属性决策的评估矩阵计算损失函数,降低了决策过程的主观性。Zhan等[]将传统三支决策中的等价关系替换为优胜关系,提出了一种新的三支决策方法,并将其应用在多属性问题的求解中。Wang等8将决策者的心理状态加入决策过程中,并以此提出了基于前景理论的三支决策模型。Li等[9]提出根据决策者对风险的偏好来计算三支决策的阈值和条件概率,以此给出了基于决策风险偏好的三支决策模型。钱进等[10]利用概念层次树构建了多层次决策表,并以此提出了多粒度层次序贯三支决策模型

集成学习是机器学习中常用的一种重要方法,其主要思想是将多个弱分类器集成到一起,以此创建一个强分类器,从而提高分类模型的精度。……

登录APP查看全文