You work for a bank and are building a random forest model for fraud detection. You have a dataset that includes transactions, of which 1% are identified as fraudulent. Which data transformation strat

Question

Options

• AModify the target variable using the Box-Cox transformation.
• BZ-normalize all the numeric features.
• COversample the fraudulent transaction 10 times.
• DLog transform all numeric features.

• A
  5% (3)
• B
  16% (10)
• C
  76% (47)
• D
  3% (2)

Explanation

Fraud datasets are highly imbalanced - 1% positive class means the model can achieve 99% accuracy by always predicting 'not fraud,' never learning the fraud signal. Oversampling the minority (fraudulent) class 10 times brings the ratio closer to 10:1, giving the model substantially more exposure to fraud patterns during training and improving its ability to detect them. This directly addresses the root cause: class imbalance. Options B (Z-normalize) and D (log transform) are feature-scaling techniques that improve numeric conditioning but do not address imbalance. Option A (Box-Cox) is a variance-stabilizing transformation for the target variable, inappropriate here because the target is binary (0/1 fraud label).

No community discussion yet for this question.