面向混合数据的代价敏感三支决策边界域分类方法

PDF下载HTML

Title:

Classification method of cost-sensitive three-way decision boundary region for hybrid data

作者:

周阳阳¹, 钱文彬^1,2, 王映龙¹, 彭莉莎³, 曾武序¹

1. 江西农业大学 计算机与信息工程学院，江西 南昌 330045;
2. 江西农业大学 软件学院，江西 南昌 330045;
3. 南京大学 工程管理学院，江苏 南京 210046

Author(s):

ZHOU Yangyang¹, QIAN Wenbin^1,2, WANG Yinglong¹, PENG Lisha³, ZENG Wuxu¹

1. School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China;
2. School of Software, Jiangxi Agricultural University, Nanchang 330045, China;
3. School of Engineering Management, Nanjing University, Nanjing 210046, China

关键词:

三支决策; 粒计算; 代价敏感; 混合数据; 正域约简; 边界域样本处理; 粗糙集; 核属性

Keywords:

three-way decisions; granular computing; cost sensitive; hybrid data; positive domain reduction; boundary region sample processing; rough set; core attribute

分类号:

TP391

DOI:

10.11992/tis.202012048

摘要:

针对现有三支决策模型的研究对象多为单一性数据的决策系统，对于混合数据边界域样本处理的研究相对较少，本文面向混合数据提出了基于核属性的代价敏感三支决策边界域分类方法。该方法基于正域约简计算混合邻域决策系统的核属性集，在此基础上计算混合邻域类，并利用三支决策规则分别将对象划分到各决策类的正域、边界域和负域；提出了一种基于代价敏感学习的三支决策边界域分类方法，并构造了误分类代价的计算方法，以此划分边界域中的对象。通过对UCI上的10个数据集进行实验对比与分析，进一步验证了本文方法，为处理边界域样本提供了一种可行有效的方法。

Abstract:

The research objects of existing three-way decisions models are mostly decision-making systems with single data. Relatively few studies on the boundary region sample processing of mixed data have been conducted. To address this issue, a classification method of a cost-sensitive three-way decision boundary region based on core attributes for hybrid data is proposed in this study. This method computes the core attribute set of the hybrid neighborhood decision system based on positive domain reduction. On this basis, the hybrid neighborhood class is calculated, and the objects are divided into the positive, boundary, and negative regions of each decision-making class through three-way decision rules. The classification method of the three-way decision boundary region based on cost-sensitive learning is proposed. Then, a calculation method of the misclassification cost is constructed to divide the objects in the boundary region. Experiments and analyses are performed on 10 datasets of UCI, which show the feasibility and the effectiveness of the proposed method for the processing of boundary region samples.