CLASSIFICATION OF CARDIAC HEART DISEASE USING REDUCED CHAOS FEATURES AND 1-NORM LINEAR PROGRAMMING EXTREME LEARNING MACHINE
Ram Sewak Singh, Barjinder Singh Saini, Ramesh Kumar Sunkaria · International Journal for Multiscale Computational Engineering · 2018
Objective: This article presents a novel method for binary classification of cardiac heart disease using generalized discriminant analysis (GDA) and the 1-norm linear programming extreme learning machine (1-NLPELM). The GDA reduces the dimension of extracted features from heart rate variability (HRV) signals, and can improve the classification accuracy by selecting best discerning features. The 1-NLPELM approach leads to a spares model depiction, in which several components of optimal solution vector will become zero and thus, without scarifying the validation accuracy, the choice of a decision function can be made using a reduced number of hidden nodes. Methodology: For classification of cardiac heart disease, the HRV signals were obtained from a standard database of healthy young (YNG), healthy elderly (ELY), normal sinus rhythm (NSR), congestive heart failure (CHF), and coronary artery disease (CAD) subjects. Initially nine features were extracted from HRV time-series signals by using chaos investigation methods, such as correlation dimension (CD), detrended fluctuation analysis (DFA) as ?1 and ?2, approximate entropy (ApEn), the results of the Poincare plot as SD1/SD2 ratio, Hurst exponent (HE), permutation entropy (PE), improved multiscale permutation entropy (IMPE) and cumulative bi-correlation (CBC). The nine features were then reduced to one feature by the GDA estimator having radial basis function (RBF), Gaussian and polynomial nonlinear kernel. This reduced feature was applied to the 1-NLPELM classifier. Results: Numerical experiments were performed on the combined database sets as YNG-ELY, NSR-CAD, NSR-CHF, CHF-CAD, YNG-CAD, CHF-YNG, ELY-CAD, and ELY-CHF subjects. The proposed method produced 98.38% validation accuracy for the NSR-CAD dataset and 100% validation accuracy for remaining datasets. Results were compared with support vector machine (SVM), extreme learning machine (ELM), differential evolution extreme learning machine (DE-ELM), online sequential extreme learning machine (OS-ELM), linear programming extreme learning machine (LPELM), 1-NLPELM, and linear discriminant analysis (LDA) + 1-NLPELM.