ABSTRACT
Causal analysis is at the heart of descriptive analytics. To perform this analysis, one will need a couple of tools: a knowledge engineering tool to capture knowledge about the domain of interest, and Bayesian networks to codify and represent that knowledge. The knowledge engineering tool and the Bayesian network tool work hand-in-hand, so they will be detailed as such. Define, Structure, Verify, Elicit (DSEV) is a field-proven method for developing Bayesian models strictly from expert knowledge with a unique blend of soft and hard skills. DSEV objective is an exhaustive probabilistic model of how the generator can fail—with expert-based pathways through each failure mode. DSEV starts, stops, and cycles with experts. Critical to the understanding of Bayesian networks is the concept of omnidirectional inference: the path is open in both directions. Relationships between variables can be modeled using any of three basic structures, individually or in combination: indirect connection; common cause; or common effect.
