A company is interested in building a fraud detection model. Currently, the data scientist does not have a sufficient amount of information due to the low number of fraud cases. Which method is MOST…

Question

Options

• AOversampling using bootstrapping
• BUndersampling
• COversampling using SMOTE
• DClass weight adjustment

• A
  3% (1)
• B
  9% (3)
• C
  74% (25)
• D
  15% (5)

Explanation

SMOTE (Synthetic Minority Oversampling Technique) generates new, synthetic fraud examples by interpolating between existing minority-class samples, giving the model richer and more varied fraud patterns to learn from - making it the strongest choice for maximizing valid fraud detection.

Why the distractors fall short:

• A (Bootstrapping/oversampling): Simply resampling existing fraud cases with replacement just duplicates data - the model memorizes the same few examples rather than learning generalizable fraud patterns, leading to overfitting.
• B (Undersampling): Discards legitimate transaction data to balance classes, which throws away potentially valuable signal and reduces overall model accuracy - the opposite of what you want with scarce data.
• D (Class weight adjustment): Penalizes misclassifying fraud more heavily during training, which helps but doesn't add new information - it reweights existing patterns rather than enriching the minority class representation.

Memory tip: Think of SMOTE as a "fraud factory" - it creates new fraud examples that look realistic by blending existing ones, rather than just copying them (bootstrapping) or shrinking the majority (undersampling). When data is scarce, synthesize, don't just sample.

No community discussion yet for this question.