自适应特征选择加权k子凸包分类

doi:10.6040/j.issn.1672-3961.0.2017.415

摘要/Abstract

摘要：

针对问题维数的增加以及不同特征对分类的作用往往不一样,导致k子凸包分类性能降低等问题,设计自适应特征选择加权k子凸包分类方法。根据传统凸包距离存在的不足引入加权k子凸包距离,在测试样本的k邻域内引入距离度量学习技术和正则化技术进行自适应的特征选择,并将自适应特征选择无缝嵌入加权k子凸包优化模型中,这样就能为不同的测试样本在不同的类别中学习自适应特征空间,得到有效的加权k子凸包距离计算方法。试验结果表明,该方法不仅能够进行降维,而且具有明显的分类性能优势。

关键词: 加权k子凸包, 度量学习, 正则化, 特征选择, 自适应

Abstract:

Because of the increase of the dimension of the problem and the effect of different features on classifier, the performance of the k sub-convex-hull classifier was seriously reduced. An adaptive feature selection weighted k sub-convex-hull classifier was designed (AWCH). A weighted k sub-convex-hull classifier was designed according to the shortcomings of conventional convex hull distance. By applying the distance metric learning and regularization technique in the k neighborhood of the test sample, an adaptive feature selection method was designed and seamlessly integrated into the optimization model on the weighted k sub-convex-hull. Through these efforts, for different test samples, an adaptive feature space in different categories could be extracted, and a valid weighted k sub-convex-hull distance could be obtained. Experimental results showed that the AWCH not only reduced the dimension of the problem, but also was significantly superior to similar classifiers.

Key words: weighted k sub-convex-hull classifier, distance metric learning, regularization, feature selection, adaptive

中图分类号:

TP391

牟廉明. 自适应特征选择加权k子凸包分类[J]. 山东大学学报 (工学版), 2018, 48(5): 32-37.

Lianming MOU. Weighted k sub-convex-hull classifier based on adaptive feature selection[J]. Journal of Shandong University(Engineering Science), 2018, 48(5): 32-37.

图/表 2

表1

表2

分类错误率和标准偏差比较"

数据集	CKNN		kCH		RLHC		AWCH
数据集	Err	Std	Err	Std	Err	Std	Err	Std
artificial	0.433 0	0.004 4	0.429 8	0.003 2	0.377 4	0.003 1	0.322 6	0.002 4
austra	0.163 5	0.006 8	0.155 1	0.005 7	0.156 1	0.005 3	0.135 0	0.004 7
balance-scale	0.110 7	0.005 7	0.103 4	0.005 0	0.103 7	0.007 7	0.067 5	0.0043
BCI	0.445 5	0.016 0	0.460 5	0.014 7	0.390 5	0.020 0	0.338 8	0.007 3
breast-w	0.032 2	0.001 5	0.031 9	0.000 9	0.026 8	0.0028	0.020 0	0.001 2
bupa	0.368 4	0.011 6	0.357 1	0.009 0	0.333 6	0.016 5	0.284 3	0.014 9
clean1	0.247 1	0.009 8	0.140 2	0.009 3	0.183 2	0.010 7	0.082 8	0.008 5
diabetes	0.260 9	0.007 7	0.259 8	0.008 1	0.257 6	0.007 9	0.202 0	0.007 4
digits	0.023 5	0.001 7	0.021 2	0.001 5	0.009 5	0.000 7	0.007 7	0.000 6
ecoli	0.145 6	0.004 9	0.146 0	0.007 8	0.132 4	0.009 1	0.127 7	0.001 3
ethn	0.046 7	0.001 9	0.033 9	0.001 6	0.021 7	0.001 5	0.010 3	0.001 4
glass	0.375 3	0.009 8	0.329 9	0.016 5	0.308 9	0.015 5	0.268 1	0.011 8
heart-statlog	0.180 0	0.006 8	0.180 4	0.005 0	0.210 7	0.019 5	0.111 5	0.017 6
ionosphere	0.167 8	0.007 4	0.161 6	0.005 9	0.105 3	0.008 2	0.078 4	0.007 4
iris	0.049 3	0.005 8	0.044 0	0.003 4	0.0427	0.004 7	0.040 7	0.004 7
isolet_norm	0.006 2	0.001 9	0.008 0	0.001 9	0.00 43	0.001 4	0.003 8	0.001 2
letter	0.063 0	0.001 0	0.060 9	0.001 1	0.034 5	0.000 6	0.028 3	0.000 4
LIBRASMovement	0.446 5	0.025 9	0.460 5	0.017 4	0.125 1	0.009 4	0.116 7	0.008 8
machine	0.233 9	0.014 5	0.233 7	0.014 7	0.088 3	0.011 5	0.083 0	0.010 6
mfeat-factors	0.047 1	0.001 3	0.044 5	0.001 2	0.030 4	0.001 5	0.030 0	0.001 1
mfeat-fourier	0.188 7	0.004 6	0.190 8	0.002 8	0.170 1	0.004 0	0.169 8	0.002 1
mfeat-karhunen	0.052 5	0.001 6	0.050 2	0.001 2	0.026 1	0.002 8	0.022 5	0.001 1
mfeat-zernike	0.185 5	0.006 7	0.179 1	0.005 0	0.165 4	0.004 6	0.149 4	0.003 7
musk	0.247 1	0.009 8	0.140 2	0.008 3	0.103 2	0.010 7	0.082 8	0.008 5
optdigits	0.017 1	0.000 5	0.016 2	0.000 4	0.008 1	0.000 4	0.008 0	0.000 3
page-blocks	0.050 8	0.000 9	0.049 4	0.001 1	0.042 4	0.002 4	0.034 7	0.000 9
Yale_32x32_face	0.486 2	0.013 9	0.475 1	0.024 3	0.278 2	0.011 6	0.229 1	0.010 0
Yale_64x64_face	0.363 1	0.017 7	0.385 3	0.016 5	0.271 3	0.008 7	0.254 0	0.008 1
ORL_32x32_face	0.354 5	0.016 8	0.367 0	0.016 1	0.034 3	0.005 8	0.030 3	0.003 3
ORL_64x64_face	0.360 3	0.012 7	0.362 5	0.012 1	0.034 8	0.006 2	0.030 2	0.005 0

表2

参考文献 20

1	VINCENT P , BENGIO Y . K-local hyperplane and convex distance nearest neighbor algorithms[J]. In Advances in Neural Information Processing Systems, 2002, 14 (1): 985- 992.
2	ZHANG Y, TANG Z M, LI Y P, et al. Ensemble learning and optimizing KNN method for speaker recognition[C]//Proceedings of the Fourth International Conference on Fuzzy System and Knowledge Discovery (FSKD). Haikou, China: [S.l.], 2007: 285-289.
3	YANG T , KECMAN V . Adaptive local hyperplane classification[J]. Neurocomputing, 2008, (71): 3001- 3004.
4	YANG T , KECMAN V . Face recognition with adaptive local hyperplane algorithm[J]. Pattern Anal Applic, 2010, (13): 79- 83.
5	WEN Guihua , JIANG Lijun , WEN Jun , et al. Perceptual relativity-based local hyperplane classification[J]. Neurocomputing, 2012, (97): 155- 163.
6	XU Jie , YANG Jian , LAI Zhihui . K-local hyperplane distance nearest neighbor classifier oriented local discriminant analysis[J]. Information Sciences, 2013, 232, 11- 26. doi: 10.1016/j.ins.2012.12.045
7	牟廉明. k子凸包分类[J]. 山西大学学报(自然科学版), 2011, 34 (3): 374- 380.
	MOU Lianming . A k sub-convex-hull classifier[J]. Journal of Shanxi University (Nat Sci Ed), 2011, 34 (3): 374- 380.
8	牟廉明. 选择性自适应k子凸包分类方法[J]. 南京大学学报(自然科学), 2013, 49 (4): 410- 416.
	MOU Lianming . Selective adaptive k sub-convex-hullclassifier[J]. Journal of Nanjing University(Natural Sciences), 2013, 49 (4): 410- 416.
9	李素姝, 王士同, 李滔. 基于LS-SVM与模糊补准则的特征选择方法[J]. 山东大学学报(工学版), 2017, 47 (3): 34- 42.
	LI Sushu , WANG Shitong , LI Tao . A feature selection method based on LS-SVM and fuzzy supplementary criterion[J]. Journal of Shandong University(Engineering Science), 2017, 47 (3): 34- 42.
10	王法波, 许信顺. 文本分类中一种新的特征选择方法[J]. 山东大学学报(工学版), 2017, 40 (4): 8- 11, 18.
	WANG Fabo , XU Xinshun . A new feature selection method for text categorization[J]. Journal of Shandong University(Engineering Science), 2017, 40 (4): 8- 11, 18.
11	戴平, 李宁. 一种基于SVM的快速特征选择方法[J]. 山东大学学报(工学版), 2017, 40 (5): 60- 65.
	DAI Ping , LI Ning . A fast SVM-based feature selection method[J]. Journal of Shandong University (Engineering Science), 2017, 40 (5): 60- 65.
12	赵佳, 王士同. 特征加权距离的半监督模糊子空间聚类算法[J]. 小型微型计算机系统, 2017, 38 (2): 405- 410.
	ZHAO Jia , WANG Shitong . Semi-supervised fuzzy subspace clustering algorithm based on feature weighted distance[J]. Journal of Chinese Computer Systems, 2017, 38 (2): 405- 410.
13	贾隆嘉, 孙铁利, 杨凤芹, 等. 基于类空间密度的文本分类特征加权算法[J]. 吉林大学学报(信息科学版), 2017, 35 (1): 92- 97. doi: 10.3969/j.issn.1671-5896.2017.01.015
	JIA Longjia , SUN Tieli , YANG Fengqin , et al. Class space density based weighting scheme for automated text categorization[J]. Journal of Jilin University(Information Science Edition), 2017, 35 (1): 92- 97. doi: 10.3969/j.issn.1671-5896.2017.01.015
14	KILIAN Weinberger . Distance metric learning for large margin nearest neighbor classification[J]. Journal of Machine Learning Research, 2009, 10, 207- 244.
15	BAR-HILLEL A , HERTZ T , SHENTAL N , et al. Learning a mahalanobis metric from equivalence constraints[J]. Journal of Machine Learning Research, 2006, 6 (1): 937- 965.
16	钱强, 陈松灿. 基于矩阵正态分布似然比测试的矩阵度量学习算法[J]. 山东大学学报(工学版), 2017, 42 (6): 37- 42.
	QIAN Qiang , CHEN Songcan . Matrix metric learning algorithm based on likelihood ratio test with matrix normal distribution[J]. Journal of Shandong University(Engineering Science), 2017, 42 (6): 37- 42.
17	郭文, 游思思, 高君宇, 等. 深度相对度量学习的视觉跟踪[J]. 中国科学:信息科学, 2018, 48 (1): 60- 78.
	GUO Wen , YOU Sisi , GAO Junyu , et al. Deep relative metric learning for visual tracking[J]. Scientia Sinica (Informationis), 2018, 48 (1): 60- 78.
18	酆勇, 熊庆宇, 石为人, 等. 深度非线性度量学习在说话人确认中的应用[J]. 声学学报, 2018, 43 (1): 112- 120.
	FENG Yong , XIONG Qingyu , SHI Weiren , et al. Deep nonlinear metric learning for speaker verification[J]. Acta Acustica, 2018, 43 (1): 112- 120.
19	CAI Deng, HE Xiaofei. Face databases[EB/OL]. (2015-03-15)[2017-05-10]. http://www.zjucadcg.cn/dengcai/Data/FaceData.html.
20	ASUNCION A, NEWMAN D J. UCI machine learning repository[EB/OL]. (1995-03-01)[2017-05-10]. http://www.ics.uci.edu/~mlearn/MLR-epository.html.

多维度评价

Viewed

Full text

Abstract

Cited

Shared

Discussed

[1]	唐杰烽,张佳,龙锦益. 基于全局冗余最小的快速多标签特征选择方法[J]. 山东大学学报 (工学版), 2025, 55(6): 21-34.
[2]	周前,李群,朱丹丹,李仪博. 基于M³C自适应虚拟惯量的海上低频风电系统协调惯量响应控制[J]. 山东大学学报 (工学版), 2025, 55(5): 30-39.
[3]	李晓辉,刘小飞,孙炜桐,赵毅,董媛,靳引利. 基于车辆与无人机协同的巡检任务分配与路径规划算法[J]. 山东大学学报 (工学版), 2025, 55(5): 101-109.
[4]	郑晓,陈鹤,周东傲,宫永顺. 基于视频描述增强和双流特征融合的视频异常检测方法[J]. 山东大学学报 (工学版), 2025, 55(5): 110-119.
[5]	高君健,廖祝华,刘毅志,赵肄江. 基于分层多智能体强化学习的个性化与信号控制联合路径引导方法[J]. 山东大学学报 (工学版), 2025, 55(3): 34-45.
[6]	刁振宇,韩小凡,张承宇,聂慧佳,赵秀阳,牛冬梅. 基于实例判别与特征增强的单图三维模型检索[J]. 山东大学学报 (工学版), 2025, 55(2): 71-77.
[7]	吴正健,吾尔尼沙·买买提,杨耀威,阿力木江·艾沙,库尔班·吾布力. 基于DRCoALTP的印刷体文档图像多文种识别方法[J]. 山东大学学报 (工学版), 2025, 55(1): 51-57.
[8]	张梦雨,何振学,赵晓君,王浩然,肖利民,王翔. 基于AMSChOA的MPRM电路面积优化[J]. 山东大学学报 (工学版), 2024, 54(6): 147-155.
[9]	王辰龑,刘轩,超木日力格. 自适应的并行天牛须优化算法[J]. 山东大学学报 (工学版), 2024, 54(5): 74-80.
[10]	方世超,滕旭阳,王子南,陈晗,仇兆炀,毕美华. 基于自适应掩码和生成式修复的图像隐私保护技术[J]. 山东大学学报 (工学版), 2024, 54(5): 111-121.
[11]	王旭晴,魏伟波,杨光宇,宋金涛,吕婷,潘振宽. 基于算法展开的图像盲去模糊深度学习网络[J]. 山东大学学报 (工学版), 2023, 53(6): 35-46.
[12]	李颖,王建坤. 基于监督图正则化和信息融合的轻度认知障碍分类方法[J]. 山东大学学报 (工学版), 2023, 53(4): 65-73.
[13]	刘财辉,周琪,叶晓文. 一种基于改进ReliefF算法的入侵检测模型[J]. 山东大学学报 (工学版), 2023, 53(2): 1-10.
[14]	刘子一,崔超然,孟凡安,林培光. 基于批归一化统计量的无源多领域自适应方法[J]. 山东大学学报 (工学版), 2023, 53(2): 102-108.
[15]	刘丁菠,刘学艳,于东然,杨博,李伟. 面向小样本目标检测任务的自适应特征重构算法[J]. 山东大学学报 (工学版), 2022, 52(6): 115-122.