Hierarchical mixtures of naive Bayes classifiers
Marco Wiering · Utrecht University Repository (Utrecht University) · 2002
Naive Bayes classifiers tend to perform very well on a large number of problem domains, although their representation power is quite limited compared to more sophisticated machine learning algorithms. In this paper we study combining multiple naive Bayes classifiers by using the hierarchical mixtures of experts system. This system, which we call hierarchical mixtures of naive Bayes classifiers, is compared to a simple naive Bayes classifier and to using bagging and boosting for combining multiple classifiers. Results on 19 data sets from the UCI repository indicate that the hierarchical mixtures architecture in general outperforms the other methods.