Combining prioritized decisions in classification

Abstract

In this paper we present an alternative evidential method of combining prioritized decisions, in order to arrive at a "consensus", or aggregate, decision. Previous studies have suggested that, in some classification domains, the better performance can be achieved through combining the first and second decisions from each evidence source. However, it is easy to illustrate the fact that going further down a decision list, to give longer preferred decisions, can provide the alternative to the method of combining only the first one and second decisions. Our objective here is to examine the theoretical aspect of an alternative method in terms of quartet - how extending a decision list of any length by one extra preferred decision affects classification results. We also present the experimental results to demonstrate the effectiveness of our alternative method. © Springer-Verlag Berlin Heidelberg 2007.