Feature Selection and Weighing for Case-based Reasoning System using Random Forests

Case-based reasoning has become a successful technique that uses the previous experience as a problem-solving paradigm. It adapts or reuses the solutions of a similar problem to solve a new one. In a case-based reasoning system, it is important to have a good similarity retrieval algorithm to retrieve the most similar cases to the query case. However, we also note that in a medical domain with increased use of electronic health records, the availability of patient cases and the related attributes have increased. Thus, as a preprocessing step or as part of the retrieval algorithm, it becomes critical to select the most informative features to improve the retrieval efficiency and accuracy in a case-based reasoning system. In this paper, we explore random forest, a popular method in machine learning, for feature selection and weighting in a case-based reasoning system and investigate the case retrieval accuracy.