An axiomatic approach for combining evidence from a variety of sources

Abstract

Development of intelligent decision-making systems which operate in complex environments like an automated factory, needs effective procedures for combining evidence from a variety of sources. Currently there are many ad-hoc evidence combination formulae as well as those based on specific methodologies such as the Bayesian, likelihood, etc. However, the limitations of these approaches have been well documented. In this paper, we begin with the testable and desirable properties required in a special problem. These properties are termed as axioms. We then systematically develop various families of combination rules obeying subsets of these axioms. The existing combination formulae form a subset of the rules mentioned here. We discuss the appropriateness of the different families of combination formulae.