A case retrieval approach using similarity and association knowledge

Abstract

Retrieval is often considered the most important phase in Case-Based Reasoning (CBR), since it lays the foundation for overall performance of CBR systems. Retrieval in CBR aims to retrieve relevant cases that can be successfully used for solving a new problem. To realize retrieval, CBR systems typically rely on a strategy that exploits similarity knowledge, and it is called similarity-based retrieval (SBR). In SBR, similarity knowledge approximates the usefulness of cases for solving a new problem. In this paper, we show that association analysis of stored cases can be used to strengthen SBR. We present a new approach for extracting and representing association knowledge from the cases using association rule mining. We propose a novel retrieval strategy USIMSCAR that qualitatively enhances SBR by leveraging both similarity and association knowledge. We demonstrate the significant advantages of using USIMSCAR over SBR through an experimental evaluation using medical datasets. © 2011 Springer-Verlag.