Predictive Parity

A fairness metric requiring that positive predictions have equal accuracy across demographic groups.

Predictive parity is a fairness metric that asks whether positive predictions mean the same thing across groups. When the model says 'yes,' it should be equally likely to be correct regardless of who the person is. This metric is also called calibration and ensures that an approval or positive classification carries equivalent meaning across demographic groups. Predictive parity is mathematically incompatible with equalized odds when base rates differ between groups, forcing practitioners to choose which type of fairness matters most for their use case.

Also known as