IEEE 3198-2025

IEEE 3198-2025, the IEEE Standard for Evaluation Method of Machine Learning Fairness, is a technical standard that specifies a method for evaluating the fairness of machine learning (ML) systems. ML systems can produce unfair outcomes for certain groups of people, with serious consequences in areas such as hiring, lending and public services. However, organisations often lack a consistent way to test for unfairness. As a result, fairness evaluations can be hard to compare or repeat. The standard provides a common, structured approach so that fairness can be assessed consistently across systems and organisations. It was published by the IEEE Standards Association in May 2025.

The standard first recognises that unfairness in ML has multiple causes, and it sorts these causes into categories. This helps evaluators understand where unfairness may arise in a system. The standard then presents the widely recognised and used definitions of ML fairness, which describe different ways a system can be considered fair.

For each definition, the standard specifies corresponding metrics and explains how to calculate them. It also provides test cases setting out detailed conditions and procedures for setting up fairness evaluations. Together, these give organisations a repeatable process for testing ML systems for fairness. Developers, testers and auditors can apply the same method across different systems and compare the results.

Auto-discovered on 2026-08-05 by OECD Catalogue Automation