Naive Bayes as an imputation tool for classification problems

We investigate the use of the naive Bayes classifier as an imputation tool for classification problems, elaborating on why the usually employed majority method may insert biases in a classification context. Considering Rubin's typology for the distribution of missingness, we have performed experiments that illustrate how an imputation process may influence classification tasks. Our results show that imputations performed by the naive Bayes can be useful for other classifiers (decision trees and nearest neighbors). In this sense, interesting hybrid systems to classify datasets with missing values can be derived.