IEEE 3198-2025

This document specifies a method for evaluating the fairness of machine learning. Multiple causes contribute to the unfairness of machine learning. In this document, these causes of machine learning unfairness are categorized. The widely recognized and used definitions of machine learning fairness are presented. This document also specifies various metrics corresponding to the definitions and how to calculate the metrics. Test cases in this document give detailed conditions and procedures to set up the tests for evaluating machine learning fairness.

The purpose of this document is to help stakeholders of machine learning systems gain a better understanding of the fairness aspects of AI systems. Using this document, the stakeholders can test and obtain quantitative values for different fairness metrics, which can be used to measure whether a machine learning system achieves the intended fairness requirements.

New IEEE Standard - Active. A method for evaluating the fairness of machine learning is specified in this standard. Multiple causes contribute to the unfairness of machine learning. These causes of machine learning unfairness are categorized. The widely recognized and used definitions of machine learning fairness are presented. Various metrics corresponding to the definitions, and how to calculate the metrics are specified in this standard. Detailed conditions and procedures to set up the tests for evaluating machine learning fairness are given by the test cases in this document.