Guiding Case-Base Maintenance: Competence and Performance?

The fundamental knowledge container in case-based reasoning is the case base of prior experiences. An important focus of recent CBR research is on maintenance strategies for achieving compact, competent case bases, as a way to improve the performance of CBR systems. However, the actual tradeoos between competence, compactness, and performance may be complex. Consequently, this paper argues for guiding the choice of case-base contents according to direct predictions of the performance eeects of retaining particular cases. The paper begins by examining the relationship between competence and performance, considering the goals and constraints that should guide addition and deletion of cases. It next proposes two performance-based metrics for guiding case addition and deletion. It then presents empirical studies of the performance tradeoos in case-base compression, and demonstrates the promise of the performance-based maintenance approach, especially for tasks with nonuniform problem distributions.