Advancing construction of conditional probability tables of Bayesian networks with ranked nodes method

Abstract

System models based on Bayesian networks (BNs) are widely applied in different areas. This paper facilitates the use of such models by advancing the ranked nodes method (RNM) for constructing conditional probability tables (CPTs) of BNs by expert elicitation. In RNM, the CPT of a child node is generated using a function known as the weight expression and weights of parent nodes that are elicited from the expert. However, there is a lack of exact guidelines for eliciting these parameters which complicates the use of RNM. To mitigate this issue, this paper introduces a novel framework for supporting the RNM parameter elicitation. First, the expert assesses the two most probable states of the child node in scenarios that correspond to extreme states of the parent nodes. Then, a feasible weight expression and a feasible weight set are computationally determined. Finally, the expert selects weight values from this set.