A deeper kind of change
Concept drift is a change in the relationship between inputs and the target. The same input now maps to a different correct output. Inputs may look identical while the underlying rule the model learned no longer holds.
Why it is harder than data drift
Data drift can be spotted from inputs alone. Concept drift usually needs labels because the inputs may be unchanged. A spam filter faces concept drift when senders invent new tricks that old patterns cannot catch.
Patterns of drift
- Sudden drift, an abrupt shift such as a policy change.
- Gradual drift, an old concept slowly replaced by a new one.
- Recurring drift, seasonal patterns that return.
Detecting it
- Track a rolling error rate and alarm when it climbs beyond control limits.
- Methods like DDM and ADWIN watch the error stream and signal change points.
- Maintain a recent labeled holdout to evaluate against fresh truth.
Responding
Retraining on recent data, or weighting recent samples more heavily, lets the model relearn the new mapping.
Key idea
Concept drift changes the input to output relationship itself, usually needs labels to detect, and is handled by tracking error over time then retraining on recent data.