Sensitivity analysis in discrete Bayesian networks

摘要: This paper presents an efficient computational method for performing sensitivity analysis in discrete Bayesian networks. The exploits the structure of conditional probabilities a target node given evidence. First, set parameters which is relevant to calculation identified. Next, this reduced by removing those combinations either contradict available evidence or are incompatible. Finally, using canonical components associated with resulting subset parameters, desired obtained. In way, important saving calculations achieved. proposed can also be used compute exact upper and lower bounds probabilities, hence easily performed. Examples illustrate methodology.