Sparse Methods in Classification and Regression

Abstract

The regression problem with a large number of variables appears in various fields of science, sparse methods make this problem more interpretable and more precise. In this work we present the method Elastic Net, which outperforms the Lasso in some situations. The elastic net have the grouping effect, while lasso does not, this is that strongly correlated predictors tend to "behave" in the same way. The lasso does not work well when the number of predictors is much grater than the number of observations, p>n. However, elastic net is useful in this situation.