When the number of features is large relative to the number of observations, or when features are correlated, OLS can overfit. regularization adds a penalty to the loss function:
