Comparing dependent combination rules under the belief classifier fusion framework

Abstract

Data fusion, within the evidence theory framework, consists of obtaining a unique belief function by the combination of several belief functions induced from various information sources. Considerable attention has been paid to combination rules dealing with beliefs induced from non-distinct information sources. The most popular fusion rule is the cautious conjunctive rule, proposed by Denœux. This latter has the empty set, called also the conflict, as an absorbing element. In fact, the mass assigned to the conflict tends toward 1 when applying a high number of the cautious conjunctive operator, and consequently, the conflict loses its initial role as an alarm signal announcing that there is a kind of disagreement between sources. This problem has led to the introduction of the normalized cautious rule which totally ignores the mass assigned to the conflict. An intermediate rule between the cautious conjunctive and the normalized cautious rules, named the cautious Combination With Adaptive Conflict (cautious CWAC), has been proposed to preserve the initial alarm role of the conflict. Despite this diversification, no great effort has been devoted until now to find out the most convenient combination rule. Thus, in this paper, we suggest to evaluate and compare the cautious conjunctive, the normalized cautious and the cautious CWAC rules in order to pick out the most appropriate one within the classifier fusion framework.