Sequential diagnosis in the independence Bayesian framework

Abstract

We present a new approach to test selection in sequential diagnosis (or classification) in the independence Bayesian framework that resembles the hypothetico-deductive approach to test selection used by doctors. In spite of its relative simplicity in comparison with previous models of hypothetico-deductive reasoning, the approach retains the advantage that the relevance of a selected test can be explained in strategic terms. We also examine possible approaches to the problem of deciding when there is sufficient evidence to discontinue testing, and thus avoid the risks and costs associated with unnecessary tests.