Which drift scenario is data drift?

• A. If input distributions change over time but the input-target relation remains constant, it is data drift.
• B. If the input-target relation changes, it is data drift.
• C. If neither data nor relation changes, it is data drift.
• D. If input data becomes larger in size, it is data drift.

Answer: (A)

Data drift happens when the distribution of the input features changes over time while the relationship between inputs and the target stays the same. In this scenario, the model keeps using the same rule to map features to the outcome, but the features it sees in production come from a different distribution than what it was trained on. That mismatch can cause predictions to degrade even though the underlying input-target mapping hasn’t changed. The statement describes this exactly: input distributions change over time but the input-target relationship remains constant. By contrast, if the input-target relationship changes, that’s concept drift; if nothing changes, there’s no drift; and simply having more data doesn’t count as drift since drift is about distribution or relationship changes, not volume.