How does data drift differ from concept drift in production data mining systems?

Data drift is change in input distribution; concept drift is change in the relationship between input and output.

Data drift refers to statistical changes in feature distributions, such as shifts in customer demographics or sensor readings. Concept drift occurs when the underlying mapping between inputs and outputs changes, such as fraud patterns evolving. Data drift may or may not affect model performance, while concept drift almost always degrades predictive accuracy. Monitoring systems must detect both using statistical tests and performance tracking.