Case-Base Maintenance of Case-Based Reasoning Systems in Classification Domains

The focus of case-based reasoning research has shifted from issues of building systems towards maintenance of those systems. However, the maintenance research in the past only considered removing cases for restoring the prediction accuracy of a case base. In consequence, knowledge may be lost, and prediction accuracy together with storage space are identified as the only requirements that are useful to satisfy customers. The purpose of this book is to show ways to improve cases, rather than only removing them, through the development and exemplary evaluation of a methodology for maintaining the case base of case-based reasoning systems in classification domains within a structured and systematic shell that indirectly measures and monitors customer requirements and restores the requirements if necessary.

Customer requirements are informal descriptions and not directly measurable. Performance measures are the formal specifications of what the customer requires from the system. For example, one customer requirement is that a customer wants the correct answer to problems, and one possible corresponding performance measure is prediction accuracy. However, performance measures can only be calculated when the system is in use, and they do not indicate any conflicts within or between cases.

To overcome this shortcoming, the methodology presented in this book introduces case properties and modify operators. Case properties indicate conflicts within or between cases. The degrees of these properties are monitored. If necessary, modify operators edit the conflicting cases to eliminate the indicated conflicts.