EHR data preparation for case based reasoning construction

Abstract

Case Based Reasoning (CBR) is the first choice in experience-based problems as diagnosis. However, building a case base for CBR is a challenging. Electronic Health Record (EHR) data can provide a starting point for building case base, but it needs a set of preprocessing steps. In this paper, we propose a case-base preparation framework for CBR systems. This framework consists of three main phases including data preparation, fuzzification, and coding. This paper will focus only on the data-preprocessing phase to prepare the EHR database as a knowledge source for CBR cases. It will use many machine-learning algorithms for feature selection and weighing, normalization, and others. As a case study, we will apply these algorithms on diabetes diagnosis data set. To check the effect of data preparation steps, a CBR prototype will being designed for diabetes diagnosis and prediction of its complications as kidney failure. The results show an enhancement to the case retrieval process of the implemented CBR system.