How does Naïve Bayes behave under feature redundancy and multicollinearity?

Feature redundancy causes Naïve Bayes to double-count evidence, leading to overconfident predictions.

When features are highly correlated, Naïve Bayes multiplies their likelihoods independently, effectively amplifying their combined effect. This violates independence assumption and leads to inflated posterior probabilities. However, classification accuracy may still remain stable if redundancy is consistent across classes.