Learning from failure: A methodology for the retrieve stage of a cardiovascular case-based reasoning system

Abstract

Finding the most suitable cases is a critical process in implementing a Case-Based Reasoning system. A poor choice of the selected case has a negative impact on all the subsequent steps in the Case-Based Reasoning cycle. Over the last years, several studies have focused on the objective of defining retrieval mechanisms for improving the Similarity-Based Retrieval process. However, these works do not integrate metrics that take into account how often a case has led to successful and unsuccessful solutions. In some areas, such as cardiovascular health, a solution that works for some individuals may lead to poor outcomes for others. In such situations, it is important that the cases in the case database accurately reflect the success ratio of each solution. Therefore, the retrieval process should integrate this measure. In this paper, we propose a new method for retrieving cases based on similarity and success. Thus, in addition to similarity, we establish new metrics that allow the average success of the case solutions in a case database to be taken into account.