Unsupervised feature selection based on self-representation sparse regression and local similarity preserving

Feature selection, as an indispensable method of data preprocessing, has attracted the attention of researchers. In this paper, we propose a new feature selection model called unsupervised feature selection based on self-representation sparse regression and local similarity preserving, i.e., UFSRL. Specifically, UFSRL is sparse reconstruction of the original data itself, rather than fitting low-dimensional embedding, and the manifold learning exerted on UFSRL model to preserve the local similarity of the data. Moreover, the l(2,1/2)-matrix norm has been imposed on the coefficient matrix, which make the proposed model sparse and robust to noise. In order to solve the proposed model, we design an effective iterative algorithm, and present the analysis of its convergence. Extensive experiments on eight synthetic and real-world data-sets are conducted, and the results of UFSRL compared with six corresponding feature selection algorithms. The experimental results show that UFSRL can effectively identify the feature subset with discriminative while reconstructing the data sparsely, and it is superior to some unsupervised feature selection algorithms in clustering performance.