Which of the following model risk management controls is MOST effective at detecting gradual label drift in a supervised classification model?
Select an answer to reveal the explanation.
Short Explanation and Infographic
Here's the deal — b is correct because label drift — where the relationship between features and the target variable changes over time — is best detected by comparing model predictions to actual observed outcomes through ongoing backtesting, which reveals declining calibration or discrimination. A may help remedy drift but does not detect it.
Full explanation below image
Full Explanation
B is correct because label drift — where the relationship between features and the target variable changes over time — is best detected by comparing model predictions to actual observed outcomes through ongoing backtesting, which reveals declining calibration or discrimination. A may help remedy drift but does not detect it. C tracks feature influence, not outcome alignment. D is a security control.