How does Naïve Bayes behave under non-stationary data distributions (concept drift)?

Naïve Bayes degrades under concept drift unless updated incrementally or reweighted.

Concept drift occurs when P(X,Y) changes over time. Since NB relies on historical frequency estimates, outdated statistics become misleading. Online updates, sliding windows, or exponential decay weighting are used to adapt. NB is particularly vulnerable when feature distributions shift rapidly.