AI Glossary Term
What does “Data Drift” mean?
A shift in real-world input data over time that can degrade model performance.
How Data Drift works
Example
A model trained to predict store demand before 2020 saw buying habits shift when routines changed. The inputs it now received looked different from its training data, and its forecasts got worse until it was retrained.
Why it matters
Models are trained on a snapshot of the past, but the world moves on. Monitoring for drift tells you when a model needs retraining before errors pile up.
Common questions
How is data drift different from model drift?
Data drift is a change in the inputs. Model drift usually refers to the resulting drop in performance, though the terms are often used loosely.
How do you detect drift?
By comparing statistics of live inputs with the training data and by tracking accuracy on freshly labeled samples.
What is the usual fix?
Retrain or update the model on recent data, and set alerts so changes are caught early.