Corresponding Author

Shichao Zhang

Document Type

Article

Abstract

In order to reduce dimensionality of high-dimensional data, a series of feature selection algorithms have been proposed. But these algorithms have the following disadvantages: (1) they do not fully consider the nonlinear relationship between data features (2) they do not consider the similarity between data features. To solve the above two problems, we propose an unsupervised feature selection algorithm based on local structure learning and kernel function. First, through the kernel function, we map each feature of the data to the kernel space, so that the nonlinear relationship of the data features can be fully exploited. Secondly, we apply the theory of local structure learning to the features of data, so that the similarity of data features is considered. Then we added a low rank constraint to consider the global information of the data. Finally, we add sparse learning to make feature selection. The experimental results show that the proposed algorithm has better results than the comparison methods.

Share

COinS