Short Definition
Distribution shift happens when production data differs meaningfully from the data used to train or evaluate a model.
Intuition
The model studied for one exam but is now being tested on a different kind of question.
Technical Definition
Distribution shift describes changes in input distribution, label distribution, conditional relationships, or deployment context between training and inference.
Example
A fraud model trained before a new payment method launches may fail when transaction patterns change.
Common Misunderstandings
Distribution shift can affect high-performing models suddenly.
Monitoring input data alone may miss label or concept drift.