Skip to content

Independent standards discovery and compliance platform

How we work

ISO/IEC TS 12791:2024 - Bias in Classification and Regression

This Technical Specification focuses on unwanted bias in machine learning tasks that assign categories or predict numerical values. It is useful to teams deciding how to examine bias from an early project design through later evaluation of the system in its intended context.

What you are buying

Identify the groups and outcomes relevant to the task before choosing an evaluation method. Then decide which differences need investigation, what evidence can be obtained and who can act if the findings point to an issue outside the model itself.

Frequently asked questions

Is this specification aimed at classification only?

It covers treatment of unwanted bias in both classification and regression machine learning tasks. Start by identifying the task used in your system and how its output affects decisions.

Is this the right document if we have evidence of unequal outcomes but have not investigated the cause?

It can help a team consider ways to treat unwanted bias. First document which groups and outcomes are involved so the team can judge whether an intervention helps and what trade-offs it introduces.