Multi-view Discriminative Feature Selection
Xiaobin Zhi, Jinghui Liu, Shaoru Wu · 2021 4th International Conference on Artificial Intelligence and Pattern Recognition · 2021
Linear discriminant analysis (LDA) is a popular feature extraction method. In the discriminative feature selection (DFS) algorithm, LDA can be transformed into a feature selection method by adding the row sparse regularization of the transformation matrix. However, this algorithm is only suitable for single view data. To address the problem of multiple views, in this paper, we propose a multi-view discriminative feature selection (MvDFS) method. In this framework, we introduce two redefined between-class and within-class scatter matrices for multi-view data into DFS algorithm. We also present an iterative algorithm for MvDFS. To show the effectiveness of MvDFS, we compare it with several related single-view feature selection methods on some real-world data sets.