Determining model correctness for situations of belief fusion

When analyzing hypotheses about specific situations of interest there is often a need to combine information from multiple sources. This principle belongs to information fusion in general, and is called belief fusion when the evidence is represented as belief functions. Because different situations can involve different forms of belief fusion, there is no single formal model that is suitable for analyzing every situation. It is therefore crucial to identify the most adequate fusion operator for modeling each class of situations to be analyzed. It can be challenging to determine the best belief fusion model for a specific situation, and there has been considerable confusion around this issue in the literature. In this paper we illustrate the importance of selecting a belief fusion model that adequately matches the situation to be analyzed, and propose a classification method for this purpose. A set of formal fusion operators is described to provide examples of specific models for classes of fusion situations.