An IS auditor notes the combined number of records utilized within the training, validation, and testing data sets exceeds the total number of records in the original data set. Which of the following

Question

Options

• AWhether the training, validation, and testing data sets were created in the correct order
• BWhether data leakage occurred from utilizing overlapping records in the data sets
• CWhether a sufficient number of records were utilized in the training data set
• DWhether the validation data set utilized the same number of records as the training data sets

• A
  4% (1)
• B
  75% (18)
• C
  8% (2)
• D
  13% (3)

Explanation

When the combined record count across all three splits exceeds the original dataset size, it is mathematically certain that records are shared (overlapping) between sets. This is the hallmark of data leakage (B), where the model is exposed to test or validation data during training, causing it to 'memorize' rather than generalize, leading to artificially inflated performance metrics that do not reflect real-world effectiveness. The creation order (A), adequacy of training set size (C), and parity between validation and training counts (D) are secondary concerns that only matter after the leakage risk is assessed and resolved.

No community discussion yet for this question.