Which technique effectively removes inputs by shrinking some coefficients to zero?

Which technique effectively removes inputs by shrinking some coefficients to zero?

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

Answer: (B)

L1 regularization promotes sparsity by driving some coefficients exactly to zero. When you add the absolute-value penalty to the loss, the optimization tends to shrink less important coefficients all the way to zero as the penalty strength increases. This means features with little predictive power are effectively removed from the model, resulting in a simpler, more interpretable solution and often better generalization when many inputs are irrelevant. Ridge regression, which uses an L2 penalty, shrinks all coefficients toward zero but generally does not set them exactly to zero, so all inputs remain in the model. Elastic Net combines L1 and L2 penalties, so it can produce sparse solutions as well, especially when the L1 portion is strong, but its sparsity depends on the balance between the two penalties. K-fold cross-validation is a validation technique used to assess model performance and help choose hyperparameters; it does not itself shrink coefficients or perform feature selection.

L1 regularization promotes sparsity by driving some coefficients exactly to zero. When you add the absolute-value penalty to the loss, the optimization tends to shrink less important coefficients all the way to zero as the penalty strength increases. This means features with little predictive power are effectively removed from the model, resulting in a simpler, more interpretable solution and often better generalization when many inputs are irrelevant.

Ridge regression, which uses an L2 penalty, shrinks all coefficients toward zero but generally does not set them exactly to zero, so all inputs remain in the model. Elastic Net combines L1 and L2 penalties, so it can produce sparse solutions as well, especially when the L1 portion is strong, but its sparsity depends on the balance between the two penalties. K-fold cross-validation is a validation technique used to assess model performance and help choose hyperparameters; it does not itself shrink coefficients or perform feature selection.