Concept Drift Explainer
Concept drift occurs when the relationship between inputs and the target outcome changes over time. A pattern that predicted the target during training may become weaker, stronger, or reversed in the environment where the model operates.
Changing customer behavior, new fraud tactics, medical practice, policy, climate, or market conditions can produce concept drift. The input values may look familiar even while their meaning for the prediction has changed.
Detecting concept drift often requires trustworthy outcome labels, because input distributions alone cannot prove that the input-to-target relationship changed. Delayed or incomplete labels make direct detection difficult and increase reliance on proxies.
Responses can include investigation, updated data, retraining, recalibration, rule changes, restricted use, or human review. Automatic retraining is not always safe because a sudden shift may reflect bad data, an incident, or manipulation.
Concept drift occurs when the relationship between model inputs and the outcome being predicted changes after training.
Key terms include concept drift, target, conditional relationship, ground truth, delayed label, proxy, retraining and adaptation.
The observed feature values may stay within their usual ranges.
What changes is how those features relate to the outcome the model must predict.
Direct evidence usually depends on comparing predictions with later outcomes.
Teams should distinguish a changed concept from missing data, pipeline faults, or random variation.
Explore more "Explainers"
Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.