Модель линейной регрессии с регулируемой селективностью для отбора признаков в задаче оценивания зависимостей по экспериментальным данным

A statistical approach for statement of the regression estimation problem in case of a small number of observations and a rich feature description is considered. For essential feature selection in regression estimation problem a probabilistic model is proposed, in which a structural parameter controls fundamental regressor selection. Experimental results of proposed algorithm are shown in comparison with the well-known feature selection methods (Lasso and Elastic Net).