A Hybrid Approach for Cases Classification of Medical Data

Abstract

Classification of cases has been widely applied in medicine, and it is helpful to disease diagnosis to a great extent. At present, the classification of medical cases is performed by physicians subjectively based on clinical theory and knowledge, which may hinder the diagnosis and treatment in some extent. In this paper, a hybrid classification approach (HCA) is proposed for medical data, it consists of two parts, including feature selection and classification. In feature selection, critical features are selected from the original features through linear correlation. Based on the selected features, cases are classified by C5.0 decision tree. And the proposed approach is evaluated through four medical datasets of diabetes, cardiac Single Proton Emission Computed Tomography (SPECT) images, lung cancer, and hepatitis survival for demonstration. On the four datasets, HCA shows a better construction for obviously higher classification accuracies, and it also outperforms some typical integrated classification methods.