Remembering To Forget: A Competence-Preserving Case Deletion Policy for Case-Based Reasoning Systems

Abstract

The utility problem occurs when the cost associated with searching for relevant knowledge outweighs the benefit of applying this knowledge. One common machine learning strategy for coping with this problem ensures that stored knowledge is genuinely useful, deleting any structures that do not contribute to performance in a positive sense, and essentially limiting the size of the knowledge-base. We will examine this deletion strategy in the context of casebased reasoning (CBR) systems. In CBR the impact of the utility problem is very much dependant on the size and growth of the case-base; larger case-bases mean more expensive retrieval stages, an expensive overhead in CBR systems. Traditional deletion strategies will keep performance in check (and thereby control the classical utility problem) but they may cause problems for CBR system competence. This effect is demonstrated experimentally and in reply two new deletion strategies are proposed that can take both competence and performance into consideration during deletion.