A free-to-play game company needs to predict whether a new user will become a paying customer within one year. The labeled training set contains 1,000 positive examples and 999,000 negative examples (1,000,000 total) with 200 features. A random forest achieved >99% accuracy on the training set but performs poorly on the test set. Which actions should the team take to address this problem? (Choose two.)

The problem describes an imbalanced dataset where positive examples are rare. The high training accuracy but poor test set performance indicates overfitting to the majority class (negative examples) and a failure to learn the minority class (positive examples). Creating more positive examples by duplicating and adding noise (oversampling/data augmentation) helps balance the dataset, allowing the model to learn more effectively from the minority class. Modifying the loss function to penalize false negatives more heavily addresses the business problem directly. In this context, a false negative means predicting a paying customer as non-paying, which is a missed revenue opportunity. This reweights the importance of correctly identifying the minority class. Increasing tree depth or number would likely worsen overfitting. Copying test samples into the training set is data leakage and invalidates evaluation. Penalizing false positives more heavily would further bias the model against predicting paying customers, exacerbating the problem.