Semi-naive bayesian classifier

The algorithm of the 'naive' Bayesian classifier (that assumes the independence of attributes) is extended to detect the dependencies between attributes to optimize the tradeoff between the 'non-naivety' and the reliability of approximations of probabilities.

Abstract

In the paper the algorithm of the 'naive' Bayesian classifier (that assumes the independence of attributes) is extended to detect the dependencies between attributes. The idea is to optimize the tradeoff between the 'non-naivety' and the reliability of approximations of probabilities. Experiments in four medical diagnostic problems are described. In two domains where by the experts opinion the attributes are in fact independent the semi- naive Bayesian classifier achieved the same classification accuracy as naive Bayes. In two other domains the semi-naive Bayesian classifier slightly outperformed the naive Bayesian classifier.