Naïve bayes variants in classification learning

Abstract

Naive Bayesian classifier is one of the most effective and efficient classification algorithms. The elegant simplicity and apparent accuracy of naive Bayes (NB) even when the independence assumption is violated, fosters the on-going interest in the model. This paper discusses issues on NB along with its advantages and disadvantages. We also present an overview of NB variants and provide a categorization of those methods based on four dimensions. These include manipulating the set of attributes, allowing interdependencies, employing local learning and adjusting the probabilities by numeric weights. Examples for each category are discussed based on 18 variants reviewed in this paper.