Unsupervised Feature Selection Algorithms: A Survey

The prodigious usage of features and variables are very high in most of the domains which are often unwanted and noisy. Feature selection is a method which is used to handle high dimensional data into dataset with less dimensions by eliminating most unwanted or repetitive features or attributes. Unsupervised feature selection means that the best features are selected among the large set of unlabelled data. some of the unsupervised feature selection algorithms namely, Clustering guided sparse structural learning (CGSSL), The Linked unsupervised feature selection (LUFS), Unsupervised spatial-spectral feature selection method, Unsupervised feature selection via optic diffraction principle, Joint embedded learning and sparse regression for unsupervised feature selection (JELSR). CGSSL is an iterative approach which integrates cluster analysis and sparse structural analysis and experimentally results are examined. The LUFS focuses on linked data to achieve linked information in selecting features and results are analyzed. The unsupervised spatial-spectral feature selection method bands are represented in prototype space and data are represented in pixel space and optimal features are obtained. Unsupervised feature selection method via optic diffraction principle based on the property of Fourier transform of probability density distribution. In JELSR embedding learning and sparse regression are fused and implemented. In this review paper, survey of above unsupervised feature selection algorithms are discussed.