Unsupervised Feature Selection Algorithm Based on Laplace Rank Constraint and Local Structure Preservation

Feature selection aims to select an optimal feature subset to reduce the dimension of original data, thereby solving the "dimension disaster" problem effectively. For feature selection, preserving the local manifold structure of the data is crucial, so how to learn an excellent local manifold structure has always been a research hotspot in this field. In this paper, we propose an unsupervised feature selection algorithm based on Laplace rank constraint and local structure preservation. First, we combine the locally linear embedding method with the Laplace rank constraint method to learn an outstanding similarity matrix. Secondly, the projection matrix is used to select features while preserving the similarity matrix. Furthermore, to avoid selecting redundant features, the regularization term about the redundancy of the projection matrix is used to select features with more discriminant information. In addition, the model optimization algorithm is proposed, and the model complexity is analyzed. Experiments on several public datasets show that our method can learn better manifold structural information, and select features with more discriminant information.