A graphical-method for solving a decision-analysis problem

A method for solving multistage decision analysis problems under uncertainty is proposed. The method is appropriate when the utility function can be decomposed to smaller factors and the joint probability function of the random variables also factorises to probabilities defined in smaller subsets of random variables. The paper uses the factorisations and the corresponding graphical structure of the problem to compute efficiently the expected utility at each stage. All computations are local in the sense that they involve a small number of variables. Then, using dynamic programming, the paper can identify an optimum strategy, depending on the available knowledge at the time that decisions are taken. The algorithm is illustrated by a worked example, and a comparisoin with existing approaches is included.