On the application of the bootstrap for computing confidence measures on features of induced Bayesian networks.
Nir Friedman, Moisés Goldszmidt, Abraham J. Wyner · 1999
In the context of learning Bayesian networks from data, very little work has been published on methods for assessing the quality of an induced model. This issue, however, has received a great deal of attention in the statistics literature. In this paper, we take a well-known method from statistics, Efron's Bootstrap, and examine its applicability for assessing a confidence measure on features of the learned network structure. We also compare this method to assessments based on a practical realization of the Bayesian methodology. 1 Introduction In the last decade there has been a great deal of research focused on the issue of learning Bayesian networks from data. With few exceptions, these results have concentrated on issues of computationally efficient induction methods and, more recently, on the issue of hidden variables and missing data. Very little work (but see below) has been published on methods or on a methodology for assessing the quality of an induced model. In this paper, w...