How can data drift affect model performance over time?

• A. It always improves model performance over time.
• B. It is unrelated to model deployment.
• C. Shifts in input data distributions can degrade accuracy if the model isn’t retrained or updated.
• D. It only affects data labeling quality.

Answer: (C)

Data drift occurs when the data the model encounters over time changes in distribution. If the features shift or the relationship between features and the target evolves, the patterns the model learned from historical data may no longer apply. Without retraining or updating to reflect the new data, the model’s predictions become less accurate, so performance tends to degrade. This happens in real-world scenarios like changing user behavior, market conditions, or system changes that alter the data you collect. To manage it, you monitor both feature distributions and model performance and trigger retraining or model adjustments when drift is detected. It’s not about always improving results, and while labeling quality can matter, the main effect of data drift is that the model becomes misaligned with current data unless it’s updated.