Concept Drift

A type of model drift caused by changes to the target variables that a machine learning model seeks to predict, which make the model less applicable. For example, if the definition of a factory part breakdown changes, a model trained to predict breakdowns according to the old definition will become less useful.

Added Perspectives

Concept drift occurs when the patterns the model learned no longer hold. In contrast to the data drift, the distributions (such as user demographics, frequency of words, etc.) might even remain the same. Instead, the relationships between the model inputs and outputs change.

- Elena Samuylova in Machine Learning in Production: Why You Should Care About Data and Concept Drift

(Blog)