Linear utility theory for belief functions (Q1121148)

Uncertainty situations are considered, in which data are described by a Dempster-Shafer belief function, i.e., a non-additive, but super-additive at all orders, set-function. A belief function can be interpreted as a lower probability, the lower envelope of the family of all probability measures which are consistent with the available data. Since belief functions form a mixture set, von Neumann-Morgenstern linear utility theory applies to belief functions, and comes up with a decision criterion which, provided a certain dominance axiom is satisfied, generalizes in a nice way the expected utility criterion. This criterion is consistent with the lower probability interpretation of belief functions and takes into account separately the decision maker's attitudes with respect to risk and with respect to ambiguity (pessimism/optimism).