An algorithm for subsequent screening non-informative variables in a linear model. (Russian) Zbl 1124.62313
Summary: An algorithm is proposed for screening non-informative variables in a linear regression model by means of alternative steps aimed at ranking variables with the help of nonsmooth regularization and deleting non-informative variables. The model with the best results on a test sample is considered as a final model. Numerical experiments show effectiveness of the algorithm proposed.
MSC:
62J05 | Linear regression; mixed models |
65C60 | Computational problems in statistics (MSC2010) |