Assessing the weighted sum algorithm for automatic generation of Probabilities in Bayesian Networks

A Bayesian Network (BN) is a probabilistic reasoning technique, which to date has been used in a broad range of applications. One of the key challenges in constructing a BN is obtaining its Conditional Probability Tables (CPTs). CPTs can be learnt from data (when available), elicited from domain experts, or a combination of both. Eliciting from domain experts provides more flexibility; however, CPTs grow in size of exponentially, thus making their elicitation process very time consuming and costly. Previous work proposed a solution to this problem using the weighted sum algorithm (WSA) [9]; however no empirical results were given on the algorithm´s elicitation reduction and prediction accuracy. Hence the aim of this paper is to present two empirical studies that assess the WSA´s efficiency and prediction accuracy. Our results show that the estimates obtained using the WSA were highly accurate and make significant reductions in elicitation.