Competence Guided Instance Selection for Case-Based Reasoning

Case-based reasoning (CBR) solves problems by reusing the solutions to similar problems stored as cases in a case-base (Kolodner, 1993); (Leake, 1996); (Smyth and Keane, 1998). For reasons of efficiency it is often desirable to be able to reduce a large case-base to a much smaller edited subset without compromising the competence of the case-base. This is obviously related to instance selection tasks but existing algorithms have been developed mainly for classification problems with discrete solution classes. Many CBR applications address non-classification tasks such as prediction and estimation tasks, planning, or design so called synthesis tasks. In this paper we describe and evaluate a number of strategies, which are unique in that they are guided by an explicit model of competence for CBR systems.