CBR Supports Decision Analysis with Uncertainty

This paper proposes a novel approach to case-based decision analysis supported by case-based reasoning (CBR). The strength of CBR is utilized for building a situation dependent decision model without complete domain knowledge. This is achieved by deriving states probabilities and general utility estimates from the case library and the subset of cases retrieved in a situation described in query. In particular, the derivation of state probabilities is realized through an information fusion process which comprises evidence (case) combination using the Dempster-Shafer theory and Bayesian probabilistic reasoning. Subsequently decision theory is applied to the decision model learnt from previous cases to identify the most promising, secured, and rational choices. In such a way we take advantage of both the strength of CBR to learn without domain knowledge and the ability of decision theory to analyze under uncertainty. We have also studied the issue of imprecise representations of utility in individual cases and explained how fuzzy decision analysis can be conducted when case specific utilities are assigned with fuzzy data.