Effectiveness of conditional possibilities on numerical information fusion

Abstract

Information fusion is a research domain that strives to establish theories that exploit and analyze the data retrieved from multiple sources. Generally, these fusion theories try to combine these data for a classification task and to make the decision efficiently. The possibility theory is one of the most known in the information fusion domain. So, the possibility distribution estimation step represents the key element of success of the fusion process based on possibilistic reasoning. In the framework of the possibility theory, we will concentrate to study the conditional possibilities distributions existing in the literature. Therefore, in this paper, we propose to present a comparative study of the different existing conditional possibilities to fuse numerical information. For this fact, we have evaluated each conditional possibility definition on 15 benchmark databases in order to deduct the best one. Thus, the experiments results provide insights that can help the researchers in the fusion information to increase the performance of its fusion/classification systems by the choosing of the most appropriate conditional possibility distribution.