Glossary term
Glossary term
Evaluation and Benchmarks
A post-prediction adjustment, typically to account for prediction bias. The adjusted predictions and probabilities should match the distribution of an observed set of labels.
Created for this library
A credit team adds a calibration layer to its boosted-tree model so predicted probabilities match the observed default rate within each risk band.
A medical diagnostics team adds a calibration layer to its deep classifier so clinicians can interpret the output as a probability rather than a raw score.
A retention team adds a Platt scaling calibration layer to its churn model so its predicted churn probability matches the actual base rate per cohort.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License