Flexible Feature Deletion: Compacting Case Bases by Selectively Compressing Case Contents

Abstract

Extensive research in case-base maintenance has studied methods for achieving compact, competent case bases. This work has examined how to achieve good solution performance while limiting the number of cases retained, using approaches such as competence-based case deletion. Two fundamental assumptions of such approaches have been (1) that cases are approximately the same size and (2) that the only way to affect case base size is by deleting or retaining entire cases. However, in some domains different cases may contain different amounts of information, causing widely varying case sizes, and case solutions may themselves be compressible, with the ability to selectively delete portions of indices or solutions while still retaining varying levels of usefulness. In accordance with this more flexible view, this paper proposes a new maintenance approach, flexible feature deletion, which removes parts of cases, enabling compression of the case base by selective—and possibly non-uniform—size reduction of individual cases. It proposes and evaluates an initial set of feature deletion strategies. Experimental results support that when cases have varying size and compressible contents, flexible feature deletion strategies may enable better system performance than case-oriented strategies for the same level of compression.