A linear model predicting real-time ad bids is overfitting and the team needs to reduce the feature set. Which approach will both reduce features and prevent overfitting?

L1 regularization (Lasso) adds a penalty proportional to the absolute value of the coefficients. This penalty can drive some coefficients to exactly zero, effectively performing feature selection by removing less important features from the model. This reduction in the number of active features directly combats overfitting. L2 regularization (Ridge) adds a penalty proportional to the square of the coefficients, shrinking them towards zero but rarely to exactly zero, so it doesn't perform feature selection. Dropout regularization is primarily used in neural networks, not typically in linear models, and randomly drops units during training to prevent co-adaptation. Retraining with more data is a general strategy to reduce overfitting but doesn't explicitly reduce the feature set.