Evolving Extended Naive Bayes Classifier

Naı̈ve Bayes classifiers are a very simple, but often ef-fective tool for classification problems, although they are based on independence assumptions that do not hold in most cases. Extended naı̈ve Bayes classifiers also rely on independence assumptions, but break them down to artifi-cial subclasses, in this way becoming more powerful than ordinary naı̈ve Bayes classifiers. Since the involved compu-tations for Bayes classifiers are basically generalised mean value calculations, they easily render themselves to incre-mental and online learning. However, for the extended naı̈ve Bayes classifiers it is necessary, to choose and con-struct the subclasses, a problem whose answer is not obvi-ous, especially in the case of online learning. In this paper we propose an evolving extended naı̈ve Bayes classifier that can learn and evolve in an online manner. 1