Which regularization uses an L1 penalty to drop inputs that are not useful?

Which regularization uses an L1 penalty to drop inputs that are not useful?

• A. Elastic Net (L1 + L2 penalties)
• B. Ridge Regression (L2 penalty)
• C. LASSO (L1 penalty)
• D. K-fold Cross-Validation

Answer: (C)

L1 regularization promotes sparsity by adding the sum of the absolute values of the coefficients to the loss. This makes some coefficients shrink all the way to zero, effectively dropping the corresponding inputs from the model. The method that uses this L1 penalty to encourage that kind of feature elimination is LASSO. Ridge regression, on the other hand, uses an L2 penalty and tends to shrink coefficients toward zero without typically making them exactly zero, so it doesn’t drop features. Elastic Net combines both L1 and L2 penalties, so it can reduce some coefficients to zero but not as aggressively as pure L1. K-fold cross-validation is a technique for estimating model performance, not a regularization penalty. Therefore, the one that drops inputs using an L1 penalty is LASSO.

L1 regularization promotes sparsity by adding the sum of the absolute values of the coefficients to the loss. This makes some coefficients shrink all the way to zero, effectively dropping the corresponding inputs from the model. The method that uses this L1 penalty to encourage that kind of feature elimination is LASSO. Ridge regression, on the other hand, uses an L2 penalty and tends to shrink coefficients toward zero without typically making them exactly zero, so it doesn’t drop features. Elastic Net combines both L1 and L2 penalties, so it can reduce some coefficients to zero but not as aggressively as pure L1. K-fold cross-validation is a technique for estimating model performance, not a regularization penalty. Therefore, the one that drops inputs using an L1 penalty is LASSO.