Bayesian Networks from Scarce Data and Expert Judgment: A Human Reliability Analysis Application

Podofillini,Luca; Dang, Vinh N.

Permalink

https://hdl.handle.net/2142/121860 Copy

Description

Title

Bayesian Networks from Scarce Data and Expert Judgment: A Human Reliability Analysis Application

Author(s)

• Podofillini,Luca
• Dang, Vinh N.

Issue Date

2023

Keyword(s)

• Bayesian belief networks
• Human reliability analysis
• Event and accident analysis
• Errors of commission
• Expert judgement

Abstract

This work addresses the development of Bayesian Belief Network (BBNs) for cases of small datasets available, emphasizing the need for a traceable development process. The application relates to a specific area of Human Reliability Analysis: the quantification of aggravating actions, as outcomes of inappropriate decisions (i.e., the so-called errors of commission, EOCs). The available data constitutes of two set of analyses of operational events (about thirty events in total) involving EOCs, in the form of patterns of analyst ratings on EOC-influencing factors and corresponding error-forcing impact (in the analysis framework of the CESA-Q method, the Quantification module of the Commission Error Search and Assessment method, developed by the authors’ research Group). This work presents a novel process for the quantification of the BBN parameters (the Conditional Probability Distributions, CPDs), combining an interpolation algorithm for populating the CPDs from part of the available evidence and Bayesian updates to adjust the BBN response to the rest of the available evidence. A first, prior BBN is developed, then sequentially updated to adjust to the two data sets. This allows some intermediate validation and puts forwards the steps for future BBN updates as new EOC events (or new analyst assessments) become available. Traceability in the BBN development comes from avoiding to elicit CPDs directly from expects, whereas, much of the judgment is constrained to the operational event analyses and to a set of anchor assessments required as input to the interpolation algorithm. These can be subject to review and, eventually, be defined such to reflect different beliefs. Once the set of inputs is given, the rest of the BBN development is performed by the algorithm, based on its pre-defined CPD filling rules and following a traceable and repeatable process.

Type of Resource

text

Language

eng

Handle URL

https://hdl.handle.net/2142/121860

Owning Collections