带拒绝域的ECOC多类分类
发布时间：2019-06-15 21:11
【摘要】：针对纠错输出编码分解框架的自身特点、从降低误判风险出发,研究了带拒绝域的ECOC多类分类方法.首先在二类划分过程中引入拒绝域,对不属于正负子类的待识别样本进行拒识;其次,在基分类器内部引入拒绝域,以最小化风险贝叶斯决策为目标,利用后验概率输出和代价矩阵寻找拒绝域阈值,对样本输出值落入拒绝域中的样本进行拒识;最后,研究了不同拒绝域输出的解码方法,并讨论了拒识码字个数和矩阵最小Hamming距离之间的关系.实验结果表明基于二类划分构造的拒绝域能够提高分类正确率,而基于基分类器构造的拒绝域能够减小分类代价.
[Abstract]:In view of the self-feature of the error correction output code decomposition framework, the ECOC multi-class classification method with the rejection domain is studied from the point of reducing the risk of misjudgment. the method comprises the following steps of: firstly, introducing a refusal domain in a second-class division process, and rejecting a sample to be identified which does not belong to the positive and negative subclasses; secondly, introducing a rejection domain into the base classifier to minimize the risk Bayesian decision as a target, and searching for a rejection domain threshold by using a posterior probability output and a cost matrix, Finally, the method of decoding the output of different denied domains is studied, and the relation between the number of rejected codes and the minimum Hamming distance of the matrix is also discussed. The experimental results show that the classification accuracy can be improved based on the refusal domain of the second-class partition structure, and the classification cost can be reduced based on the refuse field constructed by the base classifier.
【作者单位】： 空军工程大学防空反导学院;空军工程大学信息与导航学院;空军大连通信士官学校基础部;
【基金】：国家自然科学基金(No.61273275,No.61503407)
【分类号】：TP181
本文编号：2500521
[Abstract]:In view of the self-feature of the error correction output code decomposition framework, the ECOC multi-class classification method with the rejection domain is studied from the point of reducing the risk of misjudgment. the method comprises the following steps of: firstly, introducing a refusal domain in a second-class division process, and rejecting a sample to be identified which does not belong to the positive and negative subclasses; secondly, introducing a rejection domain into the base classifier to minimize the risk Bayesian decision as a target, and searching for a rejection domain threshold by using a posterior probability output and a cost matrix, Finally, the method of decoding the output of different denied domains is studied, and the relation between the number of rejected codes and the minimum Hamming distance of the matrix is also discussed. The experimental results show that the classification accuracy can be improved based on the refusal domain of the second-class partition structure, and the classification cost can be reduced based on the refuse field constructed by the base classifier.
【作者单位】： 空军工程大学防空反导学院;空军工程大学信息与导航学院;空军大连通信士官学校基础部;
【基金】：国家自然科学基金(No.61273275,No.61503407)
【分类号】：TP181
【相似文献】
相关期刊论文 前1条
1 吕佳;邓乃扬;田英杰;邵元海;杨新民;;局部学习半监督多类分类机[J];系统工程理论与实践;2013年03期
相关硕士学位论文 前3条
1 梁勃;光滑技术在多类分类问题中的应用[D];西安邮电大学;2015年
2 张大银;基于类间差异性的多类多核学习研究[D];东南大学;2015年
3 魏贵莹;基于决策粗糙集的代价敏感多类分类模型与多目标决策[D];安徽大学;2016年
，本文编号：2500521
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2500521.html
