Martin Žnidaršič (2007) Revision of probabilistic multi-criteria hierarchic models. PhD thesis.
Thesis addresses the representation of decision problems with uncertainty and the incorporation of data-based information into existing decision models. The problem are related, since the solution of one influences the solution of another. We propose an extension of the DEX decision modelling formalism, which enables modelling of uncertain rules in the utility functions of the models. The new utility functions have goal values defined as probabilistic distributions, which have parameters that describe their stability. We have adapted the procedures of decision analysis to fit this formalism. To incorporate information from new evaluated decision alternatives we propose a procedure of decision model revision, which uses background knowledge in the model and preserves it to a specified degree. For the models of extended methodology we developed a basic method of revision and some variants for improvement and use in special circumstances. The method can use the information about stability of the goal values' distributions. The experiments confirmed the usefulness of revision for decision models and pointed out some special features of particular procedures. We have presented also first practical experiences with the new formalism for the use of uncertainties in decision models.
Actions (login required)