Bayesian Classication for Set and Interval Data

Abstract

Learning naive Bayesian classiers is an impor- tant approach to probabilistic induction. However, no study has been done on naive Bayesian classiers when a query vector includes interval-valued data, and little is known about how a set of query vectors from the same unknown class can be accurately classied. In this paper, we present a new training approach to the problems above. This ap- proach is based on the \perfect aggregation" property of the Dirichlet distribution, which is usually assumed to be the prior of the variables in a Bayesian classier. The exper- imental results show that when we merge an appropriate number of query vectors with the same unknown class and the interval-valued data are formed, the acccuracies of a trained naive Bayesian classier can be promoted signi- cantly. This paper also reports a successful application of our approach in speaker recognition.