A data scientist is working on a predictive model to forecast housing prices. They notice that the model tends to overfit the training data, leading to poor performance on unseen data. To address this issue, they decide to implement regularization. Which of the following approaches would best help them reduce overfitting while maintaining model interpretability?

The data scientist should apply L1 regularization (lasso), which adds a penalty equal to the absolute size of the coefficients. By shrinking some coefficients exactly to zero, lasso reduces model complexity and eliminates noisy features, helping the model generalize to new data. Because the model still keeps a small number of non‑zero weights, it remains easy to interpret which predictors matter. For example, if a housing model originally used ten variables, lasso might keep only three, making it clear that only those three drive price predictions. This balance of reduced overfitting and clear feature importance is why lasso is the preferred choice.