ABSTRACT
A recent direction in Human Reliability Analysis (HRA) research is the development of data-informed models, which explicitly represent the quantitative and qualitative relationships among the factors influencing human performance and the error probability. In this framework, Bayesian Belief Networks (BBN) offer natural and intuitive modeling to capture complex factor relationships. However, typically BBN parameters are treated as point values, with no information about their uncertainty. This is a key shortcoming for the use of BBN in fields with limited data available (in particular HRA). In this paper, we investigate a Bayesian Learning approach to incorporate information about the uncertainty in the BBN parameter estimates. The approach is tested on a four factor HRA model of literature using artificial data. Artificial data refers to the generation of data with known properties, in order to test the modeling approach and evaluate its performance. The test shows that the approach allows capturing parameter uncertainty, propagating it through the model, and providing knowledge of the confidence bounds in the predicted HEP estimates.
