Skip to content

Stockertim Dictionary · Technology · Human resources

Ground truth

Also called ground-truth label, target label

In short

The real outcome that a model is trained to predict and is judged against; if that outcome is recorded badly or unfairly, the model learns the flaw.

  • Advanced
  • Concept
  • Checked

In detail

Every supervised machine learning model needs examples with known answers. In HR the 'answer' might be whether a past hire performed well, stayed two years or was promoted. That recorded outcome is the ground truth. The model learns which inputs predict it and is then scored on how often it matches the ground truth for cases it has not seen.

The problem is that HR outcomes are rarely neutral facts. Performance ratings reflect managers' judgements, which can carry bias; promotions reflect who was given chances; staying two years reflects how people were treated. A model trained to reproduce such labels can be highly 'accurate' while faithfully repeating past unfairness. This is why the choice of target is a design decision with ethical weight, not a technical detail.

Teams therefore ask what the label really measures, whether it is recorded the same way for every group, and whether a better outcome exists, such as an objective work-sample score. NIST's guidance on bias in AI treats the way a problem and its target are framed as one of the main sources of harmful bias.

In practice

Before building a selection model, analysts document the target outcome, how it was recorded and its known weaknesses, and they test whether the label itself differs by group. A label found to be biased is fixed or replaced before any model is trained.

Watch out

  • High accuracy against a biased label means the model is good at copying the bias.
  • Labels change meaning over time, for example when a rating scale is redesigned.

Learn it properly

Official sources

Facts last checked on 3 November 2026.

Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.

English