Adaptive graph learning for semi-supervised feature selection with redundancy minimization

Graph-based sparse feature selection plays an important role in semi-supervised feature selection. However, traditional graph-based semi-supervised sparse feature selection separates graph construction from feature selection, which may reduce the performance of model because of noises and outliers. Moreover, sparse feature selection selects features based on the learned projection matrix. Therefore, redundant features are always selected by sparse model since similar features often have similar weights, which will weaken the performance of the algorithm. To alleviate the impact of the above problems, in this study, a novel semi-supervised sparse feature selection framework is proposed, in which the quality of the similarity matrix is improved by adaptive graph learning and the negative influence of redundant features is relieved via redundancy minimization regularization. In addition, based on this framework, two specific methods are given and a unified iterative algorithm is proposed to optimize the objective function. The performance of the proposed method is evaluated by comparing it with seven advanced semi-supervised methods in terms of classification accuracy and F1 score. Extensive experiments conducted on public datasets demonstrate that the proposed methods are superior to some advanced methods. (c) 2022 Elsevier Inc. All rights reserved.

Automated summary

In this study, a novel semi-supervised sparse feature selection framework is proposed, which improves the quality of the similarity matrix through adaptive graph learning and alleviates the negative influence of redundant features through redundancy minimization regularization.