How does Naïve Bayes perform under covariate shift?

Naïve Bayes can degrade under covariate shift because feature distributions change between training and testing.

Covariate shift occurs when P(X) changes but P(Y|X) remains stable. Naïve Bayes, which models P(X|Y), may still perform reasonably if class-conditionals remain stable. However, in practice, shifts in feature distributions can distort likelihood estimates, requiring domain adaptation or reweighting techniques.