利用证据理论的多分类支持向量数据描述算法

张世醒; 韩德强; 范晓婧

doi:10.7652/xjtuxb202302016

您当前的位置：

首页 >

文章列表页 >

利用证据理论的多分类支持向量数据描述算法

—————————————————————— | 更新时间：2025-07-09

- 利用证据理论的多分类支持向量数据描述算法
- Multi-Class Support Vector Data Description Based on Evidence Theory
- 西安交通大学学报 2023年57卷第2期页码：151-160
- 作者机构：
  
  西安交通大学电信学部,西安,710049
- 作者简介：
  
  : 2022-07-11。作者简介: 张世醒(1994—),男,博士生
  韩德强(通信作者),男,教授,博士生导师。
- 基金信息：
- DOI：10.7652/xjtuxb202302016
  中图分类号： TP391
- 网络首发：2023-02-10，
  
  纸质出版：2023
- 稿件说明：
移动端阅览
张世醒, 韩德强, 范晓婧. 利用证据理论的多分类支持向量数据描述算法[J]. 西安交通大学学报, 2023,57(2):151-160.

ZHANG Shixing, HAN Deqiang, FAN Xiaojing. Multi-Class Support Vector Data Description Based on Evidence Theory[J]. 2023, 57(2): 151-160.
张世醒, 韩德强, 范晓婧. 利用证据理论的多分类支持向量数据描述算法[J]. 西安交通大学学报, 2023,57(2):151-160. DOI： 10.7652/xjtuxb202302016.

ZHANG Shixing, HAN Deqiang, FAN Xiaojing. Multi-Class Support Vector Data Description Based on Evidence Theory[J]. 2023, 57(2): 151-160. DOI： 10.7652/xjtuxb202302016.

摘要

针对原始多分类支持向量数据描述(SVDD)算法及其拓展算法忽略超球体之间的差异

且未能充分利用超球体的输出信息等问题

提出一种利用证据理论的多分类支持向量数据描述(证据SVDD多分类)算法。首先

为每一类样本训练一个超球体

并计算每个超球体的正确率与紧密程度; 接着使用上一步得到的正确率与紧密程度计算每个超球体的可靠程度; 然后

根据超球体的输出信息与可靠程度计算样本的信度函数

信度函数的生成方式采用三焦元法和基于评价矩阵的方法; 最后

根据Dempster组合规则融合上一步得到的信度函数

使用Pignistic法将融合后的信度函数转换为概率做出最终的判决。在两个人工数据集和多个UCI数据集上进行实验

结果表明

证据SVDD多分类算法相较传统算法可以获得更好的分类性能; 在多个数据集上的仿真结果表明

证据SVDD多分类算法比传统的SVDD多分类算法有3%的精度提升。

Abstract

Original multi-class support vector data description(SVDD)algorithm and its extension algorithm ignore the differences among hyperspheres and fail to make full use of the output information of hyperspheres. To address these problems

a multi-class support vector data description algorithm based on evidence theory(evidential multi-class SVDD algorithm)is proposed. Firstly

a hypersphere is trained for each class of samples

and the accuracy and closeness of each hypersphere are calculated. Secondly

the accuracy and closeness obtained in the previous step are used to calculate the reliability of hypersphere. After that

the output information and reliability of hypersphere are used to calculate the belief functions of samples

which are generated by two methods: the triple focal element method and the method based on payoff matrix. In the end

Dempster combination rule is used to fuse the belief functions

and Pignistic method is used to convert the fused belief functions into probability for final decision. Extensive experimental results on two artificial datasets and multiple UCI datasets show that the evidential multi-class SVDD algorithm achieves better classification performance than the traditional algorithm. Simulation results on multiple datasets show that the evidential multi-class SVDD algorithm has a 3% improvement in accuracy compared with the traditional multi-class SVDD algorithm.

关键词

Keywords

references

TU Junling, XU Yong, JIANG Hao, et al. Unknown fault identification method of neutron detector based on SVDD [C]//Proceedings of 2021 Chinese Intelligent Systems Conference. Singapore: Springer Singapore, 2022: 830-838.

QIU Kepeng, SONG Weihong, WANG Peng. Abnormal data detection for industrial processes using adversarial autoencoders support vector data description [J]. Measurement Science and Technology, 2022, 33(5): 055110.

闫红梅, 何明一, 梅寒雪. 一种自适应多层结构和空谱联合的高光谱图像异常检测方法 [J]. 西北工业大学学报, 2021, 39(3): 484-491.

YAN Hongmei, HE Mingyi, MEI Hanxue. Adaptive multi-layer structure with spatial-spectrum combination for hyperspectral image anomaly detection [J]. Journal of Northwestern Polytechnical University, 2021, 39(3): 484-491.

LEE K Y, KIM D W, LEE K H, et al. Density-induced support vector data description [J]. IEEE Transactions on Neural Networks, 2007, 18(1): 284-289.

刘晟, 朱玉全, 孙金津. 基于核空间相对密度的SVDD多类分类算法 [J]. 计算机应用研究, 2010, 27(5): 1694-1696.

LIU Sheng, ZHU Yuquan, SUN Jinjin. SVDD multiclass classification algorithm based on relative density in kernel space [J]. Application Research of Computers, 2010, 27(5): 1694-1696.

SUN Qiaoyan, SUN Yumei, LIU Xuejiao, et al. Study on fault diagnosis algorithm in WSN nodes based on RPCA model and SVDD for multi-class classification [J]. Cluster Computing, 2019, 22(3): 6043-6057.

TAO Xinmin, CHEN Wei, LI Xiangke, et al. The ensemble of density-sensitive SVDD classifier based on maximum soft margin for imbalanced datasets [J]. Knowledge-Based Systems, 2021, 219: 106897.

陶新民, 李晨曦, 沈微, 等. 基于密度敏感最大软间隔SVDD不均衡数据分类算法 [J]. 电子学报, 2018, 46(11): 2725-2732.

TAO Xinmin, LI Chenxi, SHEN Wei, et al. The SVDD classifier for unbalanced data based on density-sensitive and maximum soft margin [J]. Acta Electronica Sinica, 2018, 46(11): 2725-2732.

杨晨, 王婕婷, 李飞江, 等. 基于概率的支持向量数据描述方法 [J]. 计算机应用, 2019, 39(11): 3134-3139.

YANG Chen, WANG Jieting, LI Feijiang, et al. Support vector data description method based on probability [J]. Journal of Computer Applications, 2019, 39(11): 3134-3139.

AZAMI M, LARTIZIEN C, CANU S. Converting SVDD scores into probability estimates: application to outlier detection [J]. Neurocomputing, 2017, 268: 64-75.

MU Tingting, NANDI A K. Multiclass classification based on extended support vector data description [J]. IEEE Transactions on Systems, Man, and Cybernetics, 2009, 39(5): 1206-1216.

林雄, 冯海. 基于SVDD多类分类新方法的研究 [J]. 信息技术, 2008, 32(7): 100-103.

LIN Xiong, FENG Hai. Research on new method of multi-class classification based on SVDD [J]. Information Technology, 2008, 32(7): 100-103.

蔡金燕, 杜敏杰. 多分类SVDD混叠域识别新方法与故障诊断应用 [J]. 航天控制, 2012, 30(6): 83-88.

CAI Jinyan, DU Minjie. A novel approach to discriminate the overlap region of multi-class classification SVDD and fault diagnosis application [J]. Aerospace Control, 2012, 30(6): 83-88.

HAO Peiyi, LIN Y H. A new multi-class support vector machine with multi-sphere in the feature space[C]//New Trends in Applied Artificial Intelligence. Berlin: Springer, 2007: 756-765.

DEMPSTER A P. Upper and lower probabilities induced by a multivalued mapping [J]. The Annals of Mathematical Statistics, 1967, 38(2): 325-339.

SHAFER G. A mathematical theory of evidence[M]. Princeton, NJ, USA: Princeton University Press, 1976.

LIU Zhunga, PAN Quan, DEZERT J, et al. Hybrid classification system for uncertain data [J]. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2017, 47(10): 2783-2790.

HAN Deqiang, LIU Weibing, DEZERT J, et al. A novel approach to pre-extracting support vectors based on the theory of belief functions [J]. Knowledge-Based Systems, 2016, 110: 210-223.

刘卫兵, 杨艺, 韩德强. 一种基于FCOWA-ER的SVM多分类方法 [J]. 控制与决策, 2015, 30(10): 1773-1778.

LIU Weibing, YANG Yi, HAN Deqiang. A multi-class SVM based on FCOWA-ER [J]. Control and Decision, 2015, 30(10): 1773-1778.

LIU Zhunga, PAN Quan, DEZERT J. A belief classification rule for imprecise data [J]. Applied Intelligence, 2014, 40(2): 214-228.

LIU Zhunga, DEZERT J, PAN Quan, et al. A new evidential C-means clustering method [C]//2012 15th International Conference on Information Fusion. Piscataway, NJ, USA: IEEE, 2012: 239-246.

MASSON M H, DENUX T. ECM: an evidential version of the fuzzy C-means algorithm [J]. Pattern Recognition, 2008, 41(4): 1384-1397.[23] ZHAO Jianguang, LI Hongbo, ZENG Fanjing, et al. Prognostics of high frequency receiver based on evidential regression [C]//Proceedings of the IEEE 2012 Prognostics and System Health Management Conference. Piscataway, NJ, USA: IEEE, 2012: 1-5.

HE Hongshun, HAN Deqiang, YANG Yi. Cooperative semi-supervised regression algorithm based on belief functions theory [C]//2019 22th International Conference on Information Fusion. Piscataway, NJ, USA: IEEE, 2019: 1-6.

TAX D M J, DUIN R P W. Data domain description using support vectors [C]//ESANN'1999 Proceedings: European Symposium on Artificial Neural Networks. [S.l.: s.n.], 1999: 251-256.

TACNET J M, DEZERT J. Cautious OWA and evidential reasoning for decision making under uncertainty [C]//14th International Conference on Information Fusion. Piscataway, NJ, USA: IEEE, 2011: 1-8.

HAN Deqiang, DEZERT J, TACNET J M, et al. A fuzzy-cautious OWA approach with evidential reasoning [C]//2012 15th International Conference on Information Fusion. Piscataway, NJ, USA: IEEE, 2012: 278-285.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

利用证据神经网络的多分类器系统构造

最短特征线段多分类器系统设计

采用聚类特征的基本概率分配生成方法及应用

叶片生产过程的非线性轮廓控制方法研究

融合物理信息与深度算子网络的变压器绕组热-流场快速计算方法