A data scientist is building a model with thousands of features and suspects most are irrelevant. Which regularization method is most appropriate?