A shop's app starts attracting a younger audience and the model gets worse, although how age relates to buying has not changed. Which kind of drift is this?
Options
- Concept drift
- Covariate shift: the input distribution changed (Correct option)
- Data leakage
- Label shift
Answer
Covariate shift: the input distribution changed
Why
Covariate shift: the distribution of the inputs P(x) changes, while the relation P(y | x) stays the same. The model sees more of a region where it had little data and is less accurate there.
Concept drift is when P(y | x) itself changes, for example when customer behaviour changes after a price rise. Label shift is when the share of each class changes.
Covariate shift can often be detected without labels, by comparing input distributions.
Remember
Covariate shift: the inputs change. Concept drift: the rule changes.