Utility Estimation and Preference Aggregation under Uncertaintyby Maximum Entropy Inference

This paper deals with the problem how toestimate multicriteria utility functions of ordinalattributes under incomplete information. We investigatea probabilistic representation of a preference system. Bythis it is possible to communicate information about adecisionmaker’s preferences in the language ofprobabilistic conditionals, and further it enables to applythe concept of maximum entropy (ME) inference toprocess incomplete information about single multicriteriapreferences to complement and to aggregate thepreferences; both is justified by information theory.The crucial point is to understand a single utility as aspecial kind of information, and with this ME inferenceprovides utility estimates that maximize the expectedvalue of utility. Also, the here suggested access allows tohandle multi-attribute utility estimates as well. Inaddition, it is outlined how to norm the evaluated utilityfunctions, how to measure the mutual dependency oftwo sets of random variables, and how to perform anaccording analysis of the estimates’ reliability. For MEprocessingand reliability analysis the probabilisticexpert system shell SPIRIT will be used within anexample....