How do the terms non-stationarity, concept-drift and evolving data relate to each other?

Question

I often see the terms non-stationarity, concept-drift and evolving data in the same context, as if they were interchangeable. Are they? Or is there some subtle nuance that I am missing?

Answer 1

The terms above are indeed related but not similar.

A stationary process is a process whose joint distribution doesn't change over time. A typical example is white noise. Note that stationary process can be defined with a single variable (e.g., noise level). A non-stationarity process is a process in which this requirement doesn't hold.

Concept drift is defined in the framework of supervised learning. We have a concept y and it relation to the features x, change over time.

Evolving data is not a formally defined term but usually people use the term as concept drift.

You can say that concept drift cannot happen in a stationary process. However, there are non-stationarity processes that are not related to supervised learning and therefore not related to concept drift.

If you extend concept drift definition to unsupervised learning you can say the concept drift is the behaviour expected from non-stationarity processes.