Evaluating Choquet Integrals Whose Arguments are Probability Distributions

Abstract

We describe the use of the Choquet integral for finding a mean-like aggregated value of a collection of arguments with respect to a fuzzy measure. We observe the need for ordering the arguments in using this integral. We consider the case where the arguments being aggregated are random variables, probability distributions. In this case, we are faced with the problem of having to order probability distributions. Given the difficulty of obtaining a linear ordering over a collection of probability distributions, we must search for other methods for obtaining a Choquet type aggregation of a collection of probability distributions that does not require a linear ordering; we refer to these as surrogates. Here, we provide one surrogate for calculating the Choquet integral in the case where the objects being aggregated are probability distributions called the probabilistic exceedance method.