One major drawback of Naive Bayes is that it holds a strong assumption in that the features are assumed to be uncorrelated with one another, which typically is never the case.
One way to improve such an algorithm that uses Naive Bayes is by decorrelating the features so that the assumption holds true.
