You are building a logistic regression model to predict customer churn using 100 continuous numerical features. Marketing has no guidance on which features matter and wants an interpretable model showing how features affect the outcome. During training you observe a large gap between training and validation accuracy. Which two actions will (1) reduce overfitting and (2) support feature interpretability? (Choose two.)

L1 (Lasso) regularization adds a penalty proportional to the absolute value of the coefficients. This encourages some coefficients to become exactly zero, effectively performing feature selection and reducing model complexity, which combats overfitting. The non-zero coefficients then directly indicate the most important features, aiding interpretability. Recursive feature elimination (RFE) iteratively removes the least important features, retraining the model each time. This process identifies a subset of features that optimize model performance while simplifying the model, reducing overfitting, and making the remaining features more interpretable. Adding more features would likely increase overfitting. t-SNE and LDA are dimensionality reduction techniques that transform features into a new, lower-dimensional space, but they typically create new, less interpretable features, rather than selecting original ones.