Performance evaluation of information fusion systems based on belief entropy

Abstract

Information fusion systems are widely applied in many fields. However, how to quantitatively evaluate the performance of information fusion systems is still an open issue. To pioneeringly address the issue, in this paper, a performance evaluation model of information fusion systems based on Deng entropy is proposed. The proposed model quantitatively indicates the ability of evidential data fusion algorithms, including Dempster’s combination rule, average combination method, Murphy’s combination method, Yager’s combination rule, and Dubois’s combination rule, to eliminate uncertainty during the fusion process. Deng entropy serves as an indicator to characterize the uncertainty before and after fusion. We define the fusion efficiency parameter η, to numerically evaluate the performance of information fusion systems. Conflict among evidences can also be manifested in the efficiency parameter η. Several examples are presented to illustrate properties of the model. Finally, two real applications in classification are given to verify the practicability of this model.