1.School of Sciences, Liaoning Technical University, Fuxin 123000, China
2.Institute of Intelligence Engineering and Mathematics, Liaoning Technical University, Fuxin 123000, China
3.Institute of Mathematics and Systems Science, Liaoning Technical University, Fuxin 123000, China
4.Liaoning Academy of Mineral Resources Development and Utilization Technology and Equipment, Liaoning Technical University, Fuxin 123000, China
5.Fuxin Mongolian Autonomous County Vocational Education Center, Fuxin 123100, China
Show less
文章历史+
Received
Accepted
Published
2024-10-08
2025-01-04
2025-08-31
Issue Date
2025-10-30
PDF (10159K)
摘要
针对基于图的无监督特征选择算法存在挖掘数据内在信息不充分,且易受噪声干扰难以获取更具有判别性特征的问题,提出一种基于广义不相关回归和潜在表示学习的无监督特征选择方法(uncorrelated regression and latent representation for unsupervised feature selection,URLUFS)。该方法将非负矩阵分解作用于广义不相关回归模型的投影矩阵,使投影矩阵实现非线性的维数约简并获得特征选择矩阵。在特征选择矩阵的基础上,引入自适应图学习来进一步挖掘数据的局部流形结构,并对特征选择矩阵施加范数约束以保持稀疏性。利用潜在表示对数据样本间的相互关系进行学习,引导回归模型中的伪标签矩阵,从而选择出更具有判别性的特征。在8个公开的数据集上进行了数值对比实验,实验结果表明:基于广义不相关回归和潜在表示学习的无监督特征选择算法明显优于其他8种无监督特征选择算法。
Abstract
Addressing on the issues that most graph-based unsupervised feature selection algorithms often fail to fully mine the intrinsic information of data and are prone to noise interference, making it difficult to obtain more discriminative features, this study proposes an unsupervised feature selection based on generalized uncorrelated regression and latent representation learning (uncorrelated regression and latent representation for unsupervised feature selection, URLUFS). Non-negative matrix factorization is applied to the projection matrix of the generalized uncorrelated regression model to achieve nonlinear dimensionality reduction of the projection matrix and obtain the feature selection matrix. The latent representation is used to learn the mutual relationship between data samples, thereby guiding the pseudo-label matrix in the regression model. Adaptive graph learning is introduced to exploit the local manifold structure of the data, and norm constraints are imposed on the feature se-lection matrix to maintain sparsity, thereby selecting more discriminative features. Experiments are carried out on eight public datasets. The experimental results show that the proposed algorithm significantly outperforms the other eight advanced unsupervised feature selection algorithms.
基于图的无监督特征选择算法,可以使用谱分析来探索样本数据的内在结构。HE等[10]提出一种拉普拉斯得分算法(Laplacian score,LS),通过数据的局部结构计算特征的拉普拉斯得分,从而确定重要性。CAI等[11]提出一种多簇特征选择算法(multi-cluster feature selection,MCFS),通过图拉普拉斯矩阵来表示特征之间的关系,并使用回归模型来避免独立评估每个特征的贡献。NIE等[12]提出结构化最优图特征选择算法(structured optimal graph feature selection,SOGFS),通过精确连通分量来选择特征的同时捕获数据局部流行结构。ZHANG等[13]提出一种基于自适应图学习和约束的无监督特征选择算法(unsupervised feature selection via adaptive graph learning and constraint,EGCFS),将相似矩阵的构造嵌入到优化过程中,并与最大化类间散度矩阵结合来进行特征选择。YUAN等[14]提出具有自适应图的凸非负矩阵分解的无监督特征选择算法(convex non-negative matrix factorization with adaptive graph constraint,CNAFS)。与上述两种利用欧氏距离构造矩阵的方法不同,CNAFS采用最大熵原理进行图学习。ZHU等[15]提出结合图学习和L2,0范数约束的无监督特征选择(graph learning unsupervised feature selection,GLUFS),设计了一种迭代方法来求解L2,0范数约束的模型,性能较好。这些基于图的无监督特征选择算法采用投影矩阵来保留局部结构,而原始数据的投影矩阵中包含大量的噪声,会导致算法性能降低。
非负矩阵分解[16](non-negative matrix factorization,NMF)作为降维和特征提取的方法被广泛应用于数据分析。盛超等[17]提出基于子空间学习和伪标签回归的无监督特征选择,ZHOU等[18]提出基于软标签引导非负矩阵分解的无监督特征选择,两者均利用NMF,将子空间学习与回归模型结合进行特征选择。后者与广义不相关回归模型相结合,构建软标签进一步引导回归模型,提高整体模型性能。CUI等[19]提出一种子空间学习引导凸NMF的无监督特征选择,将凸NMF和多子空间结构学习相结合来选择特征。潜在表示学习作为一种新兴的方法,用来捕获数据的潜在信息,也就是样本间的聚类结构,它可以通过对称非负矩阵分解来实现[20]。TANG等[21]提出通过潜在表示学习和流形正则化的无监督特征选择(latent representation learning and graph based manifold regularization,LRLMR)。该算法将潜在表示学习应用到无监督特征选择上,与流行正则化相结合进行特征选择。XIE等[22]提出无监督特征选择的图和潜在表示的联合学习(joint learning of graph and latent representation for unsupervised feature selection,JGLUFS),将自适应图的构造集成于基于潜在表示学习的特征选择方法中,减少外部条件对图质量的影响。潜在表示学习更注重样本间的信息,同时基于潜在表示的算法大都与普通的回归模型相结合,导致在特征选择过程中无法考虑特征之间的不相关性。
JAHANIM S, AGHAMOLLAEIG, EFTEKHARIM, et al. Unsupervised feature selection guided by orthogonal representation of feature space[J]. Neurocomputing, 2023, 516: 61-76.
[2]
WANGR, ZHANGC Y, BIANJ T, et al. Sparse and flexible projections for unsupervised feature selection[J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(6): 6362-6375.
[3]
SOLORIO-FERNÁNDEZS, CARRASCO-OCHOAJ A, MARTÍNEZ-TRINIDADJ F. Filter unsupervised spectral feature selection method for mixed data based on a new feature correlation measure[J]. Neurocomputing, 2024, 571: 127111.
[4]
SHANGR H, KONGJ R, WANGL J, et al. Unsupervised feature selection via discrete spectral clustering and feature weights[J]. Neurocomputing, 2023, 517: 106-117.
MATian, LIFanhui, XIRuntao,et al.Semi-supervised image enhancement method based ongenerative adversarial network combined with transformer[J].Journal of Xi'an University of Science and Technology,2023,43(6):1207-1218.
[11]
ZHOUQ, WANGQ Q, GAOQ X, et al. Unsupervised discriminative feature selection via contrastive graph learning[J]. IEEE Transactions on Image Processing, 2024, 33: 972-986.
[12]
HEX F, CAID, NIYOGIP. Laplacian score for feature selection[J]. Advances in Neural Information Processing Systems, 2005, 18:507-514.
[13]
CAID, ZHANGC Y, HEX F. Unsupervised feature selection for multi-cluster data[C]//Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. July 25-28, 2010, Washington. New York: ACM, 2010: 333-342.
[14]
NIEF P, ZHUW, LIX L. Unsupervised feature selection with structured graph optimization[C]// Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI-16). Washington: AAAI. 2016, 30(1): 1302- 1308.
[15]
ZHANGR, ZHANGY X, LIX L. Unsupervised feature selection via adaptive graph learning and constraint[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(3): 1355-1362.
[16]
YUANA H, YOUM B, HED J, et al. Convex non-negative matrix factorization with adaptive graph for unsupervised feature selection[J]. IEEE Transactions on Cybernetics, 2022, 52(6): 5522-5534.
[17]
ZHUP C, HOUX, TANGK K, et al. Unsupervised feature selection through combining graph learning and 𝓁 2,0 -norm constraint[J]. Information Sciences, 2023, 622: 68-82.
[18]
LEED D, SEUNGH S. Algorithms for non-negative matrix factorization[C]// Proceedings of the 14th International Conference on Neural Information Processing Systems. Cambridge, Massachusetts: MIT Press. 2000, 13:556-562.
SHENGChao, SONGPeng, ZHENGWenming, et al. Subspace learning and virtual label regression based unsupervised feature selection[J]. Journal of Signal Processing, 2021, 37(9): 1701-1708.
[21]
ZHOUS X, SONGP, SONGZ H, et al. Soft-label guided non-negative matrix factorization for unsupervised feature selection[J]. Expert Systems with Applications, 2023, 216: 119468.
CAOLangcai, LINXiaochang, SUSixing. Unsupervised feature selection based on matrix factorization and adaptive graph[J]. Systems Engineering and Electronics, 2021, 43(8): 2197-2208.
[31]
SHANGR H, WANGL J, SHANGF H, et al. Dual space latent representation learning for unsupervised feature selection[J]. Pattern Recognition, 2021, 114: 107873.
[32]
SABERI-MOVAHEDF, ROSTAMIM, BERAHMANDK, et al. Dual Regularized Unsupervised Feature Selection Based on Matrix Factorization and Minimum Redundancy with application in gene selection[J]. Knowledge-Based Systems, 2022, 256: 109884.
[33]
PAPADIMITRIOUC H, STEIGLITZK. Combinatorial optimization: algorithms and complexity[M].Mineola, New York: Dover Publications,INC,1998:30-35.
[34]
STREHLA, GHOSHJ. Cluster ensembles-a knowledge reuse framework for combining multiple partitions[J]. Journal of Machine Learning Research, 2002,3: 583-617.
[35]
SHANGR H, XUK M, JIAOL C. Subspace learning for unsupervised feature selection via adaptive structure learning and rank approximation[J]. Neurocomputing, 2020, 413: 72-84.