You built an XGBoost model to predict whether a customer will return a purchase. Only 5% of examples are returns. You must maximize detection of returned items while working with a small compute budget. What is the most cost-effective hyperparameter tuning approach?

The correct approach is to use AMT to tune classweight and scaleposweight and optimize for validation:f1. Since only 5% of examples are returns, the dataset is imbalanced. classweight and scaleposweight are crucial hyperparameters in XGBoost for addressing class imbalance, helping the model pay more attention to the minority class (returns). Tuning only these specific hyperparameters is more cost-effective than tuning all possible hyperparameters, especially with a small compute budget, as it reduces the search space significantly. Optimizing for validation:f1 (F1-score) is appropriate for imbalanced datasets because it considers both precision and recall, providing a better measure of performance than accuracy, which can be misleading when one class heavily outweighs the other. Minimizing F1-score would lead to a poorly performing model.