Which regression technique uses ℓ₁ norm as its regularization term?

Lasso regression is the technique that utilizes the ℓ₁ norm as its regularization term. This method adds a penalty equal to the absolute value of the magnitude of coefficients to the loss function. The effect of this penalty is twofold: it helps to prevent overfitting by discouraging overly complex models and also has the unique ability to shrink some coefficients to exactly zero. This means Lasso regression can perform variable selection automatically, which is beneficial when dealing with datasets that have a large number of features.

In contrast, Ridge regression employs the ℓ₂ norm, which penalizes the sum of the squares of the coefficients, leading to different shrinkage behavior where coefficients are not typically reduced to zero. Linear regression, on the other hand, makes no use of regularization and focuses solely on fitting the data without any penalty, which can lead to overfitting, especially in complex datasets. Logistic regression, while also a regression technique, is focused on binary classification and does not specifically involve the ℓ₁ norm for regularization, although it can incorporate it in certain contexts, such as Lasso logistic regression. Thus, Lasso regression is the correct choice due to its specific use of the ℓ₁ norm for regularization purposes.