Case-based learning in a bipolar possibilistic framework
Jürgen Beringer, Eyke Hüllermeier · International Journal of Intelligent Systems · 2008
The paper develops a method for case-based learning and prediction within the framework of possibility theory. To this end, a possibilistic version of the similarity-guided extrapolation principle underlying the case-based learning paradigm is proposed. This version goes beyond recent proposals along those lines in that it derives a bipolar characterization of a case-based prediction: The likelihood of each potential output is characterized in terms of both a degree of evidential support and a degree of plausibility. Bipolar possibilistic predictions of such kind are quite appealing from a knowledge representational point of view as they impart much more information than standard case-based predictions. First experimental results showing how the method performs in practice are also presented. © 2008 Wiley Periodicals, Inc.