APPROXIMATION OF BELIEF FUNCTIONS

Abstract

This paper addresses the approximation of belief functions by probability functions where the approximation is based on minimizing the Euclidean distance. First of all, we simplify this optimization problem so it becomes equivalent to a standard problem in linear algebra. For the simplified optimization problem, we provide the analytic solution. Furthermore, we show that for Dempster-Shafer belief the simplified optimization problem is equivalent to the original one.In terms of semantics, we compare the approximation of belief functions to various alternative approaches, e.g. pignistic transformation for Dempster-Shafer belief and Shapley value for fuzzy belief functions. For the later one, we give an example where the approximation method has some obvious statistical advantages.Additionally, for the approximation of additive belief functions, we can provide a semantical justification.