带拒绝域的ECOC多类分类

doi:10.3969/j.issn.0372-2112.2017.11.027

电子学报 ›› 2017, Vol. 45 ›› Issue (11): 2779-2786.DOI: 10.3969/j.issn.0372-2112.2017.11.027

带拒绝域的ECOC多类分类

雷蕾¹, 王晓丹¹, 罗玺², 王玮³

1. 空军工程大学防空反导学院, 陕西西安 710051;
2. 空军工程大学信息与导航学院陕西西安 710077;
3. 空军大连通信士官学校基础部, 辽宁大连 116600

收稿日期:2016-05-24 修回日期:2016-10-10 出版日期:2017-11-25
- 作者简介:
- 雷蕾,女,1988年生于四川南充,博士生.研究方向为智能信息处理和目标识别.E-mail:wendyandpaopao@163.com;王晓丹,女,1966年生于陕西汉中,教授,博士.研究方向为模式识别,机器学习等.E-mail:afeu_wang@163.com;罗玺,男,1988年生,硕士,讲师.研究方向为智能信息处理.E-mail:luoxi19887302@126.com;王玮,男,1985年,讲师.研究方向为计算机应用.
- 基金资助:
- 国家自然科学基金 (No.61273275，No.61503407）

Design of Reject Option for Multi-classification Based on ECOC

LEI Lei¹, WANG Xiao-dan¹, LUO Xi², WANG Wei³

1. The Air and Missile Defense Institute, Air Force Engineering University, Xi'an, Shaanxi 710051, China;
2. The Information and Navigation Institute, Air Force Engineering University, Xi'an, Shaanxi 710077, China;
3. The Fundamental Department, Dalian Air Force Communication NCO Academy, Dalian, Liaoning 116600, China

Received:2016-05-24 Revised:2016-10-10 Online:2017-11-25 Published:2017-11-25
- Supported by:
- National Natural Science Foundation of China (No.61273275, No.61503407)

摘要/Abstract

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.

关键词: 多类分类, 纠错输出编码, 拒绝域, 支持向量数据描述, 贝叶斯决策

Abstract: Aiming at reducing misclassification costs,this paper studies the design of reject options for ECOC multi-classification based on its properties.The first level of reject option is constructed in the process of bipartitions to recognize an instance whose real labels does not belong to the meta-subclasses.Meanwhile,the second reject rule is presented in the dichotomizers based on posterior probabilities and cost matrix to make the minimum-risk Bayesian decision.Finally,different decoding strategies are analysed according to different reject output.The relationship between the number of rejected positions and the minimum Hamming distance of matrix is discussed.The two-stage reject rule makes the ECOC multi-classification with rejection come true and reduce the misclassification error and costs.

Key words: multi-classification, error-correcting output codes, rejection option, support vector domain description, Bayesian decision

中图分类号:

TP391

雷蕾, 王晓丹, 罗玺, 等. 带拒绝域的ECOC多类分类[J]. 电子学报, 2017, 45(11): 2779-2786.

LEI Lei, WANG Xiao-dan, LUO Xi, et al. Design of Reject Option for Multi-classification Based on ECOC[J]. Acta Electronica Sinica, 2017, 45(11): 2779-2786.

参考文献

[1] T G Dietterich,E Kong.Error correcting output codes corrects bias and variance[A].Proc of the 21th International Conference on Machine Learning[C].AAA Press,1995.313-321.
[2] T G Dietterich,G Bakiri.Solving multi-class learning problems via error-correcting output codes[J].Journal of Artificial Intelligence Research,1995,34(2):263-286.
[3] Mikel Galar,Alberto Fernández,Edurne Barrenechea,Francisco Herrera.DRCW-OVO:Distance-based relative competence weighting combination for one-vs-one strategy in multi-class problems[J].Pattern Recognition,2015,48(1):28-42.
[4] Francesco Ciompi,Oriol Pujol,Petia Radeva.ECOC-DRF:Discriminative random fields based on error correcting output codes[J].Pattern Recognition,2014,47(6):2193-2204.
[5] Zhou Jingdeng,Wang Xiaodan,et al.Research on the Unbiased Probability Estimation of Error-Correcting Output Coding[J].Pattern Recognition,2011,44(7):1552-1565.
[6] Jie Zhou,Hanchuan Peng,Ching Y.Suen.Data-driven decomposition for multi-class classification[J].Pattern Recognition,2008,41(1):67-76.
[7] Paolo Simeone,Claudio Marrocco,Francesco Tortorella.Two stage reject rule for ECOC classification systems[J].LNCS,2011,6713(1):217-226.
[8] 雷蕾,王晓丹,罗玺等.ECOC多类分类研究综述[J].电子学报,2014,42(9):1794-1800. LEI lei,WANG Xiao-dan,LUO Xi,et.al.An overview of multi-classification based on error-correcting output codes[J].Acta Electrinica Sinica,2014,42(9):1794-1800.(in Chinese)
[9] David M J Tax.Support vector data description[J].Machine Learning,2004,54(1):45-66.
[10] Bing Zhou,Yiyu Uao,Jigang Luo.Cost-sensitive three-way email spam filtering[J].Journal of Intelligent Information Systems,2014,42(1):19-45.
[11] P Simeon,C Marrocco,F Tortorella.Design of reject rules for ECOC classification systems[J].Pattern Recognition,2012,45(2):863-875.
[12] Lei LEI,Wang Xiao-dan,et al.Hierarchical error-correcting output codes based on SVDD[J].Pattern Analysis and Applications,2016,19(1):163-171.
[13] S Escalera,David M J Tax,O Pujol,P Radeva,Robert P W Duin.Subclass problem-dependent design for error-correcting output codes[J].IEEE Trans on Pattern Analysis and Machine Intelligence,2008,30(6):1041-1054.
[14] 周进登,王晓丹.基于混淆矩阵的自适应纠错输出编码多类分类方法[J].系统工程与电子技术,2012,34(7):220-226. Zhou Jindeng,Wang Xiaodan.Multiclass classification of adaptive error-correcting output codes based on confusion matrix[J].Systems Engineering and Electronics,2012,34(7):220-226.(in Chinese)
[15] Jindeng Zhou,Xiaodan Wang,et al.Coding design for error-correcting output codes based on perception[J].Optical Engineering,2012,51(5):322-331.

带拒绝域的ECOC多类分类

Design of Reject Option for Multi-classification Based on ECOC

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 10

编辑推荐

Metrics

[1]	谭晓衡, 谢朝臣, 郭坦. 基于区域感知贝叶斯决策的5G超密集异构网络联合垂直切换技术研究[J]. 电子学报, 2018, 46(3): 582-588.
[2]	雷蕾, 余晓东, 王晓丹, 罗玺, 王艺菲. 基于免疫克隆选择的最优ECOC编码输出[J]. 电子学报, 2018, 46(12): 3044-3049.
[3]	盛凯, 刘忠, 周德超, 魏启航, 冯成旭. 一种基于证据理论的多类半监督分类算法[J]. 电子学报, 2018, 46(11): 2642-2649.
[4]	杨金鸿, 邓廷权. 一种基于单簇核PCM的SVDD离群点检测方法[J]. 电子学报, 2017, 45(4): 813-819.
[5]	雷蕾, 王晓丹, 罗玺, 周进登, 陈琴. ECOC多类分类研究综述[J]. 电子学报, 2014, 42(9): 1794-1800.
[6]	周进登, 周红建, 杨云, 郭长华, 胡洪宇. 基于神经网络的纠错输出编码方法研究[J]. 电子学报, 2013, 41(6): 1114-1121.
[7]	方景龙, 王万良, 王兴起, 龙哲, 祁萌. 求解多示例问题的支持向量数据描述方法[J]. 电子学报, 2013, 41(4): 763-767.
[8]	周进登;王晓丹;权文;许燕;姚旭. 加权解码在解决纠错输出编码Consistent-Diverse平衡问题的应用[J]. 电子学报, 2011, 39(7): 1514-1522.
[9]	方景龙;王万良;梁东升;周其力. 二阶环形间隔分类器[J]. 电子学报, 2011, 39(7): 1634-1638.
[10]	李昆仑;黄厚宽;田盛丰. 模糊多类SVM模型[J]. 电子学报, 2004, 32(5): 830-832.