L1 with L2 Regularization is a method commonly used in traditional machine learning to reduce generalization errors. The following is about the two. The right way is:

Question

Options

• AL1 Regularization can do feature selection
• BL1 with L2 Regularization can be used for feature selection
• CL2 Regularization can do feature selection
• DL1 with L2 Regularization cannot be used for feature selection

• A
  81% (26)
• B
  9% (3)
• C
  6% (2)
• D
  3% (1)

Explanation

L1 regularization (Lasso) uniquely drives coefficients to exactly zero, effectively eliminating irrelevant features entirely - this sparse solution property is what makes feature selection possible.

Why the distractors are wrong:

• B is the trickiest distractor: while Elastic Net (L1 + L2) retains some sparse-solution behavior from its L1 component, the L2 term counteracts the sparsity-inducing effect and weakens feature selection - pure L1 is the definitive method. The question targets this nuance.
• C is wrong because L2 (Ridge) only shrinks coefficients toward zero but almost never reaches exactly zero, so no features are eliminated.
• D is factually flawed in the opposite direction - it overcorrects by claiming Elastic Net cannot do feature selection at all, which is too strong a claim.

Memory tip: Think of L1 as a "hard cutter" (zeros out coefficients) and L2 as a "soft shrinker" (squeezes but keeps all features). When you add L2 to L1, you dull the knife. "L1 = Lasso = Loss of irrelevant features" is a handy mnemonic to lock in why A is the clean, correct answer.

No community discussion yet for this question.