A global bank trained a churn model on 1,000 customer records, of which 100 are labeled as customers who left. After training in SageMaker Data Wrangler, the model predicts only the negative class (never predicts churn). What change to the training data will most likely fix this and produce more accurate churn predictions?

The model's failure to predict the positive class (churn) indicates a severe class imbalance problem, where the minority class (churned customers) is significantly underrepresented. With only 10% of the data representing churn, the model likely learned to always predict the majority class to minimize overall error. SMOTE addresses this by generating synthetic samples for the minority class, effectively balancing the dataset and allowing the model to learn the characteristics of churned customers. Removing outliers might improve general model performance but won't directly solve the class imbalance. Normalizing features is a good practice but doesn't address the class distribution. Undersampling the majority class could also balance the dataset, but it discards valuable information, potentially leading to a less robust model than SMOTE.