Ground Truth
Definition
Ground truth refers to the reference against which predictions or measurements are evaluated.
In machine learning, ground truth data are usually accepted target values or labels. They can come from real events, measurements, or human annotations and may themselves contain errors or uncertainties.
Simply explained
Like the answer key used for checking – only the answer key itself can also be faulty.
Why it matters in practice
Central for training, evaluation, and quality measurement of AI systems.
Example from business
A quality model predicts rejects; the later inspected incoming goods control serves as a reference for evaluation.
Limits and pitfalls
- Treat Ground Truth as infallible truth; ignore inconsistent annotations; allow data leakage between reference and model training.
Origin & history
Ground Truth is used in machine learning as a reference for evaluating models. Such reference data can originate from observed events, measurements, or human assessments.
Use in mid-sized companies
- Possible application: build verified reference datasets for document classification, quality control, or predictive models.
Frequently asked questions
Can Ground Truth be wrong?
Yes, measurement, process, and annotation errors are possible. Must it always come from humans? No, it can also arise from reliable event or measurement data.
Sources
- Ground Truth · Google Machine Learning Glossary, Ground Truth
Source check: 11.10.2026.
Editorially reviewed on: