Singular sources mining using evidential conflict analysis

Singular sources mining is essential in many applications like sensor fusion or dataset analysis. A singular source of information provides pieces of evidence that are significantly different from the majority of the other sources. In the Dempster–Shafer theory, the pieces of evidence collected by a source are summarized by basic belief assignments (bbas). In this article, we propose to mine singular sources by analyzing the conflict between their corresponding bbas. By viewing the conflict as a function of parameters called discounting rates, new developments are obtained and a criterion that weights the contribution of each bba to the conflict is introduced. The efficiency and the robustness of this criterion is demonstrated on several sets of bbas with various specificities.

► Combination of conflicting sources of information as part of the belief functions theory. ► Introduction of a new criterion to value the individual contribution of each source to the conflict. ► Introduction of a new property for conflict contribution criteria called homogeneity. ► The new criterion is the only one possessing this interesting property. ► The criterion relies notably on pairwise conflicts of sources, it is thus easy to understand the meaning behind it.