medium10 pts · first answer countscalibrationmodel-evaluation
Probabilities that cannot be trusted
Your gradient boosting model ranks customers well (good AUC), but when it says "0.8 probability", only about 50% of those customers actually convert. What should you do if the business needs reliable probabilities?
Discussion 0
Join free to post in the discussion.
Related questions
- Big gap between train and validation accuracyMachine Learning Engineer
- Why keep a separate test setMachine Learning Engineer
- Handling class imbalanceMachine Learning Engineer
- ROC-AUC after calibrationMachine Learning Engineer
- Why keep a test setData Scientist
- Which prediction hurts log loss mostData Scientist
- Validating a forecasting modelData Scientist
- Brier score for two predictionsData Scientist