Graph-regularized feature selection based on spectral learning and subspace learning

Abstract Feature selection is an important approach for reducing the dimension of high-dimensional data. In recent years, many feature selection algorithms have been proposed. However, most of them only exploit information from the data space. They often neglect useful information contained in the feature space, and typically do not exploit information about the underlying geometry of the data. To overcome these problems, we introduce new unsupervised feature selection methods based on the feature selection framework of joint embedding learning, sparse regression, and subspace learning, and extend the framework by introducing the feature graph. The framework of NSSRD comprises four main parts: dual-graph nonnegative spectral learning, dual-graph sparse regression, feature selection, and optimization. The framework of SGFS comprises three main parts: sparse subspace learning, local structure preserving, and update rules.