Michele LaurelliBlog

Concept drift

Concept drift: A change over time in the relationship between inputs and expected outcomes that ages learned behavior. Unlike a feature-only shift, it changes what the model should predict. It can be gradual, abrupt or recurring and requires temporal evaluation and retraining strategies.

In italiano: Concept DriftLegge V · Overfitting, bias e collasso dei modelli

Unlike a feature-only shift, it changes what the model should predict. It can be gradual, abrupt or recurring and requires temporal evaluation and retraining strategies.

Examples

  • New fraud patterns
  • Changing user preferences
  • Updated business rules