Which statement correctly distinguishes data drift from model drift?

• A. Data drift is a change in input data distribution; model drift is a change in the relationship between inputs and outputs; detect via monitoring performance, drift tests, and statistical checks.
• B. Data drift and model drift describe the same phenomenon.
• C. Data drift refers to model performance drop, while model drift refers to data distribution shifts.
• D. Data drift only occurs with small datasets; model drift only occurs with large datasets.

Answer: (A)

Data drift and model drift describe different ways an ML system can change its behavior over time. Data drift is a change in the input data distribution the model sees—feature values, ranges, and correlations shift from what the model was trained on. Model drift is a change in the relationship between inputs and outputs—the predictive mapping that the model learned no longer matches how the world behaves, so the same inputs lead to different outputs or worse predictions.

This distinction is why the best answer emphasizes both parts and how to detect them: watch for changes in the input data distributions with statistical checks and drift tests, and simultaneously monitor model performance and other indications that the input-output relationship has shifted. In practice, you’d look for data drift by comparing the current feature distributions to the training distributions, and you’d look for model drift by tracking performance metrics, calibration, or reliability as the environment evolves.

The other statements mix these ideas up or introduce unrelated notions. Drift is not simply the same phenomenon as data and model evolving together, it’s about separate changes in data versus the learned relationship. And drift isn’t tied to dataset size—issues can occur with both small and large datasets.