Subspace Learning and Virtual Label Regression Based Unsupervised Feature Selection

With the rapid development of information technology, a lot of unlabeled high-dimensional data are generated. To cope with these data, in this paper, we proposed a subspace learning and virtual label regression based unsupervised feature selection method. First, from the viewpoint of matrix factorization, we combined subspace learning and feature selection into a joint framework and constrained the feature selection matrix with an L_2,1-norm to select features while finding the low-dimensional representation of the original data space. Then, we utilized the regression function to learn the mapping relationship between the feature subspace and the virtual label space. With the guide of virtual label matrix and regression function, we can select the discriminative features. Finally, we introduced a graph Laplacian to explore the local information hidden in the data space and feature space. We conducted extensive experiments on six public datasets. The results show that our method is superior to some state-of-the-art unsupervised feature selection methods.