Evaluations of evidence combination rules in terms of statistical sensitivity and divergence

The theory of belief functions is one of the most important tools in information fusion and uncertainty reasoning. Dempster's rule of combination and its related modified versions are used to combine independent pieces of evidence. However, until now there is still no solid evaluation criteria and methods for these combination rules. In this paper, we look on the evidence combination as a procedure of estimation and then we propose a set of criteria to evaluate the sensitivity and divergence of different combination rules by using for reference the mean square error (MSE), the bias and the variance. Numerical examples and simulations are used to illustrate our proposed evaluation criteria. Related analyses are also provided.