A New Evidence Combination Scheme for Decision Assistant

Abstract

In D-S evidence theory, the Dempster's rule always leads to counter-intuitive results when combining conflicting evidences. The methods proposed to solve these problems all simply focus on BBA features, and ignore the decision assistant purpose. In order to get consistent results, a new combination scheme is proposed, which imitates the thinking of our brain, and fuses evidences with different conflict levels by different rules. The new scheme firstly classifies the evidences according to decision support property, then, fuses the BBAs in the same class through classical discounting method, in which the discount factor is related to probability quality, finally, combines the evidences conflicting seriously through another discount way selecting BBA reliability as the factor. The numeric examples show that the new scheme evaluates BBAs from decision assistant point all-time long, and can solve the counter-intuitive paradoxes effectively.