基于K最近邻的代价敏感三支决策边界域处理模型

发布时间：2018-01-13 09:06

本文关键词：基于K最近邻的代价敏感三支决策边界域处理模型　出处：《数码设计》2016年02期 　论文类型：期刊论文

【摘要】：三支决策理论是Yao在研究粗糙集和决策粗糙集时提出的,其主要目的是为粗糙集三个域提供合理的语义解释,即正域POS(X)、负域NEG(X)和边界域BND(X)。目前,如何有效地处理边界域已成为三支决策理论研究的热点问题。例如,基于CCA的三支决策模型提出了三种方法对边界域样本进行处理,分别是距中心最近原则、距边界最近原则和万有引力原则,但是这三种方法都没有考虑到分类问题的代价敏感性。本文在基于CCA的三支决策模型的基础上,针对边界域的处理问题,提出了一种基于K最近邻的代价敏感三支决策边界域处理模型。该模型首先根据样本分布特征寻找最优K值,然后根据与样本边界距离最小的K个覆盖的类别和代价敏感损失函数对边界域样本进行划分。实验结果表明,与基于CCA的三支决策模型中的处理方法相比,本文模型在最优K值下的分类结果的高代价样本的误分类数显著减少,分类损失更小,而且总分类错误率较低。
[Abstract]:The three-branch decision theory is put forward by Yao when studying rough set and decision rough set. Its main purpose is to provide reasonable semantic explanation for three domains of rough set, that is, positive domain POSX). At present, how to deal with the boundary domain effectively has become a hot issue in the theory of three branches of decision making. For example, how to deal with the boundary domain effectively has become a hot issue in the theory of decision making. Based on the three-branch decision model of CCA, three methods are proposed to deal with the samples in the boundary domain, namely, the nearest distance principle, the nearest distance boundary principle and the universal gravitation principle. However, these three methods do not take into account the cost sensitivity of classification problems. Based on the three-branch decision model based on CCA, this paper deals with the problem of boundary domain. A cost sensitive three-branch decision making boundary domain processing model based on K-nearest neighbor is proposed. Firstly, the optimal value of K is found according to the characteristics of sample distribution. Then, the samples of boundary domain are divided according to the classes and cost sensitive loss functions of K covers with the smallest boundary distance. The experimental results show that the proposed method is compared with that in the three-branch decision model based on CCA. Under the optimal K value, the classification error number of high-cost samples is significantly reduced, the classification loss is smaller, and the total classification error rate is lower.
【作者单位】：安徽大学计算机科学与技术学院;安徽大学计算智能与信号处理教育部重点实验室;
【基金】：国家自然科学基金项目(61673020、61602003)资助
【分类号】：TP18
【正文快照】： 引言 Yao在粗糙集和决策粗糙集研究中提出了三支决策理论,该理论将传统的正域、负域的二支决策语义拓展为正域、边界域和负域的三支决策语义[1-2]。目前的三支决策理论的研究主要是基于粗糙集的三支决策理论,而其中最具有代表性的是决策粗糙集理论模型 (Decision Theoreti

【相似文献】