Meta learning ensemble technique for diagnosis of cardiac autonomic neuropathy based on heart rate variability features

Heart Rate Variability (HRV) attributes form an important set of tests, usually collected for patients with different kinds of pathology such as diabetes, kidney disease and cardiovascular disease. The aim of this study was to examine the role of HRV attributes for improving the diagnosis of Cardiac Autonomic Neuropathy (CAN). We investigated the performance of various base classifiers for the most essentials features for CAN combined with the HRV attributes. To get the optimal subset of features, we used a feature selection method based on mean decrease accuracy (MDA), which is implemented in the Random Forest classifier. Random Forest consistently outperformed all other base classifiers. A number of ensemble classifiers have also been investigated using Random Forest to enhance the diagnosis of CAN when Ewing battery tests were combined with HRV attributes. The results improved classification accuracy compared to existing classifiers with the best results obtained by AdaBoostM and MultBoost ensembles.