Increasing precision of credible case-based inference

Abstract

Credible case-based inference (CCBI) is a new and theoretically sound inferencing mechanism for case-based systems. In this paper, we formally investigate the level of precision that CCBI-based retrieval results may yield. Building upon our theoretical findings, we derive a number of optimization criteria that can be utilized for learning such similarity measures that bring about more precise predictions when used in the scope of CCBI. Our empirical experiments support the claim that, given appropriate similarity measures, CCBI can be enforced to produce highly precise predictions while its corresponding level of confidence is only marginally impaired. © Springer-Verlag Berlin Heidelberg 2008.