Abstract

The probability assessments of a Bayesian belief network generally involve inaccuracies. These inaccuracies influence the reliability of the network's output. An integral part of investigating the reliability of output is to study robustness. Robustness pertains to the extent to which varying the probability assessments of a Bayesian belief network influences its output. Output robustness is studied by subjecting the network to a sensitivity analysis. In this paper, we address the issue of robustness of a belief network's output in view of the threshold model for decision making. We present a method for sensitivity analysis that provides for the computation of bounds between which a network's assessments can be varied without inducing a change in recommended decision.

Citations