The aims of this study are summarized in the following items: first, to investigate the class discrimination power of long-term heart rate variability (HRV) features for risk assessment in patients suffering from congestive heart failure (CHF); second, to introduce the most discriminative features of HRV to discriminate low risk patients (LRPs) and high risk patients (HRPs), and third, to examine the influence of feature dimension reduction in order to achieve desired accuracy of the classification. We analyzed two public Holter databases: 12 data of patients suffering from mild CHF (NYHA class I and II), labeled as LRPs and 32 data of patients suffering from severe CHF (NYHA class III and IV), labeled as HRPs. A K-nearest neighbor classifier was used to evaluate the performance of feature set in the classification. Moreover, to reduce the number of features as well as the overlap of the samples of two classes in feature space, we used generalized discriminant analysis (GDA) as a feature extraction method. By applying GDA to the discriminative nonlinear features, we achieved sensitivity and specificity of 100% having the least number of features. Finally, the results were compared with other similar conducted studies regarding the performance of feature selection procedure and classifier besides the number of features used in training.
Keywords: Congestive heartfailure (CHF); Generalized discriminant analysis (GDA); Heart rate variability (HRV); k-Nearest neighbor.
Copyright © 2015 Elsevier Ireland Ltd. All rights reserved.