In this paper, we propose to learn sources independence in order to choose
the appropriate type of combination rules when aggregating their beliefs. Some
combination rules are used with the assumption of their sources independence
whereas others combine beliefs of dependent sources. Therefore, the choice of
the combination rule depends on the independence of sources involved in the
combination. In this paper, we propose also a measure of independence, positive
and negative dependence to integrate in mass functions before the combinaision
with the independence assumption.