What does model drift refer to?

• A. Degradation of AI model performance over time due to changing data patterns
• B. Improvement over time due to more data
• C. A shift in software licensing terms
• D. A change in hardware temperature

Answer: (A)

Model drift is when a deployed model starts to perform worse over time because the data it encounters in production changes from what it was trained on. As patterns in input data shift—due to seasonality, evolving user behavior, new market conditions, or other factors—the relationships the model learned may no longer hold, leading to lower accuracy, miscalibrated outputs, or unfair results. This is why degradation due to changing data patterns is the best description. To illustrate, a fraud detector might miss new tactics that emerge after deployment. Addressing drift involves monitoring performance and data distributions, and retraining or updating the model as needed. The other options don’t describe performance changes caused by shifts in data.