Learning Naïve Bayes Classifiers From Attribute Value Taxonomies and Partially Specified Data

Zhang, Jun and Honavar, Vasant (2004) Learning Naïve Bayes Classifiers From Attribute Value Taxonomies and Partially Specified Data. Technical Report TR 04-03, Department of Computer Science, Iowa State University.

Abstract

Partially specified data are commonplace in many practical applications of machine learning where different instances are described at different levels of precision relative to an attribute value taxonomy (AVT). This paper describes AVT-NBL -- a variant of the Naïve Bayes Learning algorithm that effectively exploits user-supplied attribute value taxonomies to construct compact and accurate Naïve Bayes classifiers from partially specified data. Our experiments with several data sets and AVTs show that AVT-NBL yields classifiers that are substantially more accurate and more compact than those obtained using the standard Naïve Bayes learner.