Which situation MOST strongly indicates data leakage?

The correct answer is B. Using test data statistics during normalization. Data leakage occurs when information from outside the training set illegitimately influences model training, causing overly optimistic performance estimates. Using test set statistics (e.g., mean and standard deviation) to normalize the entire dataset - including training data…

NVIDIA Certified Associate (NCA) Core AI Concepts

Question

Options

AHigh training accuracy
BUsing test data statistics during normalization
CLarge model size
DSmall batch size

How the community answered

(23 responses)

A
4% (1)
B
91% (21)
D
4% (1)

Explanation

Data leakage occurs when information from outside the training set illegitimately influences model training, causing overly optimistic performance estimates. Using test set statistics (e.g., mean and standard deviation) to normalize the entire dataset - including training data - leaks future/holdout information into training. In practice, normalization parameters must be computed only on the training set and then applied to validation and test sets. Option A (high training accuracy) is a symptom of overfitting but not necessarily leakage. Options C and D (large model size, small batch size) are architectural or optimization choices unrelated to data leakage.

Topics

#Data Leakage#Machine Learning Best Practices#Data Preprocessing#Model Evaluation

Community Discussion

No community discussion yet for this question.

Full NCA-AIIO Practice