Cho-k-NN: a method for combining interacting pieces of evidence in case-based learning
Eyke Hüllermeier · 2005
The case-based learning paradigm relies upon memorizing cases in the form of successful problem solving experience, such as e.g. a pattern along with its classification in pattern recognition or a problem along with a solution in case-based reasoning. When it comes to solving a new problem, each of these cases serves as an individual piece of evidence that gives an indication of the solution to that problem. In this paper, we elaborate on issues concerning the proper combination (aggregation) of such pieces of evidence. Particularly, we argue that cases retrieved from a case library must not be considered as independent information sources, as implicitly done by most case-based learning methods. Focusing on the problem of prediction as a performance task, we propose a new inference principle that combines potentially interacting pieces of evidence by means of the so-called (discrete) Choquet-integral. Our method, called Cho-k-NN, takes interdependencies between the stored cases into account and can be seen as a generalization of weighted nearest neighbor estimation. 1