Improved risk analysis for large projects: Bayesian networks approach
Abstract
Generally risk is seen as an abstract concept which is difficult to measure. In this thesis,
we consider quantification in the broader sense by measuring risk in the context of large projects.
By improved risk measurement, it may be possible to identify and control risks in such a way that
the project is completed successfully in spite of the risks.
This thesis considers the trade-offs that may be made in project risk management,
specifically time, cost and quality. The main objective is to provide a model which addresses the
real problems and questions that project managers encounter, such as:
• If I can afford only minimal resources, how much quality is it possible to achieve?
• What resources do I need in order to achieve the highest quality possible?
• If I have limited resources and I want the highest quality, how much functionality do
I need to lose?
We propose the use of a causal risk framework that is an improvement on the traditional
modelling approaches, such as the risk register approach, and therefore contributes to better
decision making.
The approach is based on Bayesian Networks (BNs). BNs provide a framework for causal
modelling and offer a potential solution to some of the classical modelling problems. Researchers
have recently attempted to build BN models that incorporate relationships between time, cost,
quality, functionality and various process variables. This thesis analyses such BN models and as
part of a new validation study identifies their strengths and weaknesses. BNs have shown
considerable promise in addressing the aforementioned problems, but previous BN models have
not directly solved the trade-off problem. Major weaknesses are that they do not allow sensible
risk event measurement and they do not allow full trade-off analysis. The main hypothesis is that
it is possible to build BN models that overcome these limitations without compromising their
basic philosophy.
Authors
Fineman, MilijanaCollections
- Theses [3822]