Evaluation
Model drift
Model drift is a change over time that degrades or alters model behavior.
How it works
Input distributions, target relationships, or user behavior can shift after deployment. Detection requires reference windows and actionable thresholds.
A concrete example
Illustrative: positive-class prevalence rises from 8% in training to 15% this month.
Do not confuse it with Data drift
data drift is input change; model drift is the broader observed behavior change and may have several causes.