The practice of stochastic sensitivity analysis described in the decisionanalysis literature is a testimonial to the need for considering deviationsfrom precise point estimates of uncertainty. We propose the use of Bayesianfuzzy probabilities within an influence diagram computational scheme forperforming sensitivity analysis during the solution of probabilistic inferenceand decision problems. Unlike other parametric approaches, the proposed schemedoes not require resolving the problem for the varying probability pointestimates. We claim that the solution to fuzzy influence diagrams provides asmuch information as the classical point estimate approach plus additionalinformation concerning stochastic sensitivity. An example based on diagnosticdecision making in microcomputer assembly is used to illustrate this idea. Weclaim that the solution to fuzzy influence diagrams provides as muchinformation as the classical point estimate approach plus additional intervalinformation that is useful for stochastic sensitivity analysis.
展开▼