Sparse Regression Feature Selection Algorithm with Feature Variance Weighting

Aiming at the problem of insufficient discriminative feature selection ability in the existing sparse regression based feature selection algorithm, a sparse regression feature selection algorithm with feature variance weighting is proposed by introducing the variance information of features. In this method, the variances of the features are used to assist in constructing the sparse penalty term, that is, the variances of the features are used to construct the feature weights, which are added to the corresponding rows of the transformation matrix in the sparse penalty term, so as to achieve a more discriminative feature selection. At the same time, the variances of the features are also used to construct the initial regression transformation matrix, which can improve the global optimization searching capability of the algorithm. The whole model is solved using the Iteratively Re-Weighted method, which ensures the convergence of the algorithm. Comparative experiments on seven datasets have conducted to validate the effectiveness of the proposed method.