The primary issue is a severe class imbalance, where the minority class (fraud) is underrepresented, leading to poor model performance (high false negatives). The Synthetic Minority Oversampling Technique (SMOTE) is a data-level method specifically designed to address this. By generating new, synthetic examples of the minority class in the feature space, SMOTE rebalances the class distribution of the training dataset. This forces the model to learn the characteristics of the fraudulent transactions more effectively, thereby improving its ability to detect them and reducing the number of false negatives.

A. Early stopping is a regularization technique to prevent overfitting; it does not correct the underlying class imbalance that causes the model to ignore the minority class.

B. Oversampling should be applied to the minority (fraudulent) class to increase its representation. Oversampling the majority (non-fraudulent) class would exacerbate the imbalance.

C. While hyperparameter tuning can improve model performance, it is an indirect approach that does not fundamentally alter the imbalanced data distribution, which is the root cause of the problem.

1. AWS Machine Learning Blog. (2021). Handling imbalanced data with Amazon SageMaker. In the section "Data-level approach," the blog states, "The data-level approach focuses on rebalancing the data distribution so that the minority class is well-represented... A popular oversampling technique is the Synthetic Minority Oversampling Technique (SMOTE)." This directly supports using SMOTE as a preprocessing step.

2. Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligence Research, 16, 321-357. Section 2, "SMOTE: Synthetic Minority Over-sampling Technique," describes the algorithm as a method to over-sample the minority class by creating “synthetic” examples rather than by over-sampling with replacement, which directly addresses the scenario's need. DOI: https://doi.org/10.1613/jair.953

3. Amazon SageMaker Developer Guide. XGBoost Algorithm. The guide discusses the scaleposweight hyperparameter, a common technique to handle class imbalance by weighting the loss function. This highlights that while model-level solutions exist (related to option C), a direct data correction method like SMOTE (option D) is a distinct and valid preprocessing approach as requested by the question.