Repository

Fairness Metrics

Purpose

Two groups can receive positive decisions equally often and still experience different kinds of error. This Python example joins predictions, observed outcomes and group membership by user ID. Its README lays out a side-by-side comparison of selection rates, false positives, false negatives and recall.

There is a gap between the example and the code. The published function refers to undefined group variables while calculating false omissions, so it stops before returning the table.

Sources

Fairness Metrics

Fairness Metrics for Binary Classification

Published Python implementation