计算机科学与探索 ›› 2015, Vol. 9 ›› Issue (1): 94-104.DOI: 10.3778/j.issn.1673-9418.1406024

• 人工智能与模式识别 • 上一篇    下一篇

基于粗糙集的多标记决策系统知识获取方法

余  鹰1,2,3,4,苗夺谦2,3+,赵才荣2,3,王映龙4   

  1. 1. 华东交通大学 软件学院,南昌 330013
    2. 同济大学 计算机科学与技术系,上海 201804
    3. 同济大学 嵌入式系统与服务计算教育部重点实验室,上海 201804
    4. 江西农业大学 软件学院,南昌 330013
  • 出版日期:2015-01-01 发布日期:2014-12-31

Knowledge Acquisition Methods for Multi-Label Decision System Based on Rough Sets

YU Ying1,2,3,4, MIAO Duoqian2,3+, ZHAO Cairong2,3, WANG Yinglong4   

  1. 1. School of Software, East China Jiaotong University, Nanchang 330013, China
    2. Department of Computer Science and Technology, Tongji University, Shanghai 201804, China
    3. Key Laboratory of Embedded System and Service Computing, Ministry of Education, Tongji University, Shanghai 201804, China
    4. School of Software, Jiangxi Agricultural University, Nanchang 330013, China
  • Online:2015-01-01 Published:2014-12-31

摘要: 在多标记决策系统中,每个对象由单个实例进行表示,同时对应于多个决策属性。粗糙集理论已有的研究工作主要集中在单一决策系统的研究上,对于多决策系统只是简单地将它分解成多个单一决策系统。直接变换的方法忽视了决策属性之间的相关性和共现性,影响决策的精度。基于粗糙集模型,分别针对属性值为离散型和连续型的情况,提出了离散型多标记决策系统知识获取算法DML和连续型多标记决策系统知识获取算法CML。这两种算法均考虑了标记之间的相关性,在离散多标记决策系统中,采用决策链方式传递属性间的相关性,而在连续多标记决策系统中,扩展了传统粗糙集模型,重新定义了粗糙近似。实验表明,不论是离散型还是连续型决策系统,考虑决策属性之间的相关性均可以提高预测的准确率。

关键词: 粗糙集, 多标记, 决策系统, 规则提取

Abstract: Multi-label learning deals with the problem where each instance is represented by a feature vector while associated with multiple decision attributions. The existing research on rough sets focuses on decision system with single decision attribute. For the decision system with multiple decision attributes, it is simply converted into several single decision systems. One single decision system is built for one decision attribute, which neglects the correlation among the different decision attributions and reduces the classification accuracy. Based on rough sets, this paper proposes two decision-making algorithms DML and CML for discrete and continuous attributes respectively. These two algorithms consider the correlation between the labels. DML constructs a decision chain to deliver the correlation among decision attributes, while CML extends the traditional rough set model and redefines the upper and lower approximation. The experimental results show that both discrete and continuous multi-label decision systems which consider the correlation between decision attributes perform better than those algorithms which neglect the correlation among decision attributions.

Key words: rough sets, multi-label, decision system, rule extraction