What is the difference between concept drift and data drift?

Data drift (covariate shift) means the distribution of input features changes over time while the relationship between inputs and target stays the same, P(X) changes but P(Y|X) does not. Example: a new user segment shifts the age distribution of your inputs. Concept drift means the relationship between inputs and target itself changes, P(Y|X) changes, so the same input now maps to a different correct output. Example: during an economic shock, income features that once predicted low default risk no longer do. The distinction matters for diagnosis and remedy: data drift is detectable by comparing input distributions (KS test, PSI) without labels and may or may not hurt accuracy; concept drift usually requires labels or delayed feedback to detect and almost always degrades performance, forcing a retrain on recent data. Monitoring should track both input distributions and, when labels arrive, prediction quality.