Concept Drift

noun

say it: “KON-sept DRIFT” /ˈkɑnsɛpt ˈdrɪft/

A change over time in the relationship between a model's inputs and the result it predicts, which makes what it learned out of date.

What is concept drift?

Concept drift is when the link between the features and the answer changes, even if the features look the same: fraudsters change their tricks, or a hot day stops selling what it used to. The model's old lessons stop being true, and only retraining on recent answers fixes it.

In a sentence

  • “Concept drift set in when fraudsters started making many small purchases instead of a few big ones.”
  • “Retraining every month kept concept drift from wearing the model down.”

Level: Freelancer · More AI Basics terms at this level

Updated