Modeling qualitative judgements in Bayesian networks
Jose Louis Galan Caballero · Queen Mary Research Online (Queen Mary University of London) · 2008
Although Bayesian Networks (BNs) are increasingly being used to solve real world problems [47], their use is still constrained by the difficulty of constructing the node probability tables (NPTs).A key challenge is to construct relevant NPTs using the minimal amount of expert elicitation, recognising that it is rarely cost-effective to elicit complete sets of probability values.This thesis describes an approach to defining NPTs for a large class of commonly occurring nodes called ranked nodes.This approach is based on the doubly truncated Normal distribution with a central tendency that is invariably a type of a weighted function of the parent nodes.We demonstrate through two examples how to build large probability tables using the ranked nodes approach.Using this approach we are able to build the large probability tables needed to capture the complex models coming from assessing firm's risks in the safet:v or finance sector.The aim of the first example with the National Air-Traffic Services(NATS) is to show that using this approach we can model the impact of the organisational factors in avoiding mid-air aircraft collisions.The resulting model was validated by NATS and helped managers to assess the efficiency of the company handling risks and thus, control the likelihood of air-traffic incidents.In the second example, we use BN models to capture the operational risk (OpRisk) in financial institutions.The novelty of this approach is the use of causal reasoning as a means to reduce the uncertainty surrounding this type of risk.This model was validated against the Basel framework [160], which is the emerging international standard regulation governing how financial institutions assess OpRisks.This model \vas d('\'eloped with the help of the experts at the NATS.\\'ith their help we built the topology, probability tables and validated the model's predictions.The experts at NATS described the BNs model as a useful decision making tool for identifying potential safety risks and allocating resources.Chapter 7.This chapter explains the use of BN models to assess operational risk in financial institutions.The need to identif~T and measure this risk is given by the introduction of new regulatory measures by the Basel Committee to be effective by 2007.\Ye discuss the content of such regulation and the framework the Committee introduce to assess this type of risk.The aim of this chapter is to compare the current methods against BN models and to show that BNs can offer a viable alternative to them.To this end, we have built a BN model that combines data and experts' judgment in compliance with the Basel Committee requirements.By combining these information sources, we improve the accuracy of the predictions while increasing our knowledge of the domain area.Furthermore, the BN model explicitly highlights an organisation's weaknesses showing how potential risk emerge from their interaction, thus making the model an essential tool for risk management and for the regulatory body.Chapter 8.In this chapter we give a summary and we draw the conclusion of this thesis.We comment on the lessons learnt during the development of the NATS project.Appendix A As part of this thesis we develop a BN tool to help elicit and build the NPTs needed for the NATS project.This tool implements the ideas originated from this project.Ideas such as the ranked node approach.Thanks to this tool we were able to built the NPTs and to run sensitivity analysis.As part of the lessons learnt and the research study we implemented also a Bezier curve, Histograms and Normal distribution to elicit experts' opinions.These tools are now part of the AgenaRisk tool [134]. Appendix BThe Safety Attitude Questionnaire (SAQ).This appendix shows the set of questions on SAQ.:). Uncertainty of the model dnived f}'(JIll the difficultv tu find nIl possible factors that contribute on a breakdown.Using BN helps HS study the possible models bv maximizing the most probable explanation.6.It can perform "what if' scenarios to test models' predictions 2 .7. Graphical representation shows explicitly the operational process involved in C1 busi-nC'ss unit and its expected outcome.In the field of flnanee this is of particular interest.The methods, required b~' tlw financial regulCltors [164]' must be able to perform such an analysis for auditor~' reasons.Also from the point of view of risk management and decision making [123]:• Subjective estimation encourages ownership of the risk.• De-centralises responsibility: within the organisation we find different subcultures that reflect the particular environment of that part of the company.It is better for them to identify and measure their own risks rather than having one measure to "flt all".For these reason we modeled the impact of organisational culture, in the NATS project, as the result of an organisation-division culture which is how a particular set of people interpret that mainstream culture.• Better controls: it focuses in control weaknesses thus optimising the risk management.However, the use of BN leaves us with a number of challenges:1. Eliciting networks.Given that BN models are a semantic representation of the domain, we face the problem of defining the meaning of qualitative inputs and estimating the value of the quantitative ones.2. Building large NPTs.One of the biggest obstacles to building large BN models is to provide the NPTs.In some cases, these NPTs need thousands of probability estimations.To elicit each value would at best represent a time consuming and an error prone task and at worst is impossible.We study these challenges in the following chapters.2This technique is part of the Basel requirements to build risk models [162]