Model drift
A deployed model's real-world performance degrading over time because the data it sees in production has shifted away from the data it was trained on, or because the underlying relationship between input and correct answer has changed. Model monitoring is what detects it; retraining is usually what fixes it.
Why exams ask this
Tested as the reason a model that scored well at launch degrades months later with no code change. The exam distinguishes data drift, where the input distribution shifts, from concept drift, where the relationship between input and correct output itself changes.
Relevant to
Related concepts
Resources
No resources linked to this concept yet.