A Novel Machine Learning Approach for Detection of Coronary Artery Disease Using Reduced Non-linear and Chaos Features

Abstract

In this research paper, authors present an automated system in this paper that integrates a ranking technique with Principal Component Analysis (PCA), Generalized Discriminant Analysis (GDA) and a 1-Norm Bidirectional Extreme Learning Machine (1-NBELM) to reliably classify normal and coronary artery disease groups. Twenty chaotic and non-linear attributes were hauling out from the Heart Rate Variability (HRV) signal to detect coronary artery disease groups. The HRV data for this study derived from a typical database of Normal Old (ELY), Young (YNG), and Coronary Artery Disease (CAD) people. Fisher, Wilcoxon and Bhattacharya were used to compute the rankings of attributes. GDA then turned the ranking features into a new feature. The Radial Basis Function (RBF) kernel was used to transfer original features to a high-dimensional feature space in GDA and PCA, and then it was deployed to 1-NBELM, which utilized the sigmoidal or multiquadric non-linear activation. Numerical experiments were performed on the combination of database sets as Young-ELY, Healthy-CAD, and Healthy ELY-CAD subjects. The numerical results show that ROC with GDA and 1-NBELM approach achieved an accuracy of 98.12±0.14, 96.21±0.12 and 99.87±0.28 for Young-CAD, Young-ELY and Healthy ELY-CAD groups with the use of sigmoidal and multiquadric activation function. The Fisher with GDA and 1-NBELM and Bhattacharya with GDA and 1-Norm Extreme Learning Machine (1-NELM) approach achieved an accuracy of 99.98±0.21 for all databases.