Knowledge MaxxingData drift
Data drift is when the statistical distribution of a model's input data changes over time compared to what it was trained on, even if the underlying relationship to the outcome stays the same.
Prerequisite: Concept drift
What matters
Because a model was optimized for its original training distribution, a large shift in input distribution can degrade performance even without any change in what the correct outcome should be.
Example
A model trained mostly on desktop-browser traffic may perform worse if user traffic drifts to be mostly mobile, even if mobile users' true behavior patterns are similar.
Common misconception
Data drift does not necessarily mean the model's learned relationship is wrong; it means the new inputs look different from what the model was trained and tuned on.
Recall question
How does data drift differ from concept drift?