---
title: "Concept drift"
description: "A change in the relationship between a system's inputs and the correct output, so the same input should now get a different answer."
canonical: https://aigovernanceengineer.com/glossary/concept-drift
author: "Jorge García Aibar"
license: "CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)"
doi: https://doi.org/10.5281/zenodo.22956197
version: "0.5.0"
updated: 2026-09-24
---

# Concept drift

A change in the relationship between a system's inputs and the correct output, so the same input should now get a different answer [1]. Unlike data drift, it shows only in outcomes: in production it appears in performance on fresh labels and in change-point tests on the error rate.

- Developed in: [ch. 11, Contrast pairs](https://aigovernanceengineer.com/bok/ai-defined#contrast-pairs); [ch. 15, Drift: what moves and how to see it](https://aigovernanceengineer.com/bok/governing-deployment#drift-what-moves-and-how-to-see-it)
- Chapters: [ch. 11, AI Defined](https://aigovernanceengineer.com/bok/ai-defined) · [ch. 15, Deployment](https://aigovernanceengineer.com/bok/governing-deployment)
- Contrast with: [Data drift](https://aigovernanceengineer.com/glossary/data-drift)
- In the glossary chapter: https://aigovernanceengineer.com/bok/glossary#t-concept-drift

## Commonly confused

- **Data drift** vs **Concept drift**: The inputs change, against the relationship between inputs and the right answer changing The first shows in input monitors before labels arrive; the second only in outcomes on fresh labels

## Sources

[1] "Learning under Concept Drift: A Review" (IEEE TKDE 31(12); detection, understanding and adaptation; arXiv 2004.05785). Lu et al.. 2018. https://arxiv.org/abs/2004.05785 (verified: primary)