Detection of congestive heart failure based on spectral features and extreme learning machine

Abstract

In this paper we proposed a novel approach to evaluate the classification performance of features derived from various spectral investigation methods for congestive Heart Failure (CHF) analysis using ranking methods, Kernel Principal Component Analysis (KPCA) and binary classifier as 1-norm linear programming extreme learning machine (1-NLPELM). For this study, thirty different features are extracted from heart rate variability (HRV) signal by using spectral methods like multiscale Wavelet packet (MSWP), higher order spectra (HOS) and auto regression (AR) model. Top ten features were extracted by ranking methods and then reduced to only one feature by KPCA having kernel function as radial basis function (RBF) which wasfurther applied to 1-NLPELM binary classifier. For this purpose, the HRV data were taken from standard database of Normal sinus rhythm (NSR),elderly (ELY) and Congestive heart failure (CHF) subjects. Numerical experiments were being done on the combination of database sets as NSR-CHF, NSR-ELY, and ELY-CHF subjects. The numerical results show that features at third level of decomposition of HRV data sets MSWP shows lowest p-value. Thus, third level of MSWP features are better than other features extracted by auto regression (AR) model and higher order spectra (HOS) spectral methods.