How does Naïve Bayes behave in adversarial feature correlation injection attacks?

Naïve Bayes is vulnerable to adversarial correlation injection, where attackers exploit independence assumptions.

Attackers introduce correlated feature patterns that violate Naïve Bayes independence assumption. Since NB treats features independently, correlated malicious signals get over-amplified, skewing posterior probabilities. This can significantly degrade classification performance even with small perturbations in training data.