Which penalty combines L1 and L2 penalties to shrink coefficients and drop useless inputs?

Get ready for the GARP Risk and AI Exam with flashcards and multiple choice questions. Each question comes with hints and explanations. Prepare for success!

Elastic Net blends L1 and L2 regularization, so it can both shrink coefficients and drop useless inputs. The L1 part promotes sparsity, pushing some coefficients exactly to zero to perform feature selection, while the L2 part shrinks coefficients to stabilize estimates and handle correlated features. This combination is especially useful when you have many predictors or multicollinearity, as it tends to select groups of related features rather than choosing a single one arbitrarily. The overall strength and the balance between the two penalties are controlled by regularization parameters, giving a model that is both sparse and robust. The other options aren’t penalties that mix both parts: Ridge uses only L2 and won’t zero out inputs, LASSO uses only L1 and can be unstable with correlated features, and K-fold cross-validation is a resampling method, not a penalty.