Skip to content
BENIBIX Learn

Concept Drift

AI⚙️ MLOps & DeploymentProduction Monitoring

Also known as Model Drift, Label Drift

The right answer for the same input changes. So a model's rules go out of date.

Think of it like

Think of it like the rules of a game changing. The same move that won last year now loses.

Example

Lucy's fraud model learned that big payments at night were risky. Scammers switched to many small payments by day, and the model missed them though the inputs looked normal.

Why it matters

It breaks a model even when the inputs look fine. So a check on inputs alone can miss it.

Definition

Concept drift is when the link between inputs and the right answer changes. Then the rules a model learned stop being true. The inputs can look just the same as before, but the right output for them is now different. Fraud is a common case, since people who cheat change their tricks once they are caught. How people shop after a big event is another. You can only see concept drift clearly by comparing the model's answers with the real results as they come in.
Last year's line split good from bad. This year the same line no longer fits.

Common mistake

Watching only the inputs. Concept drift can leave the inputs the same, so check answers against real results too.

Key takeaways

  1. 1
    The rules change. Not just the inputs.
  2. 2
    Inputs may look the same. Answers differ.
  3. 3
    Check real outcomes. To see it.

Compare