Improved Classification of Phonocardiography Signal Using Optimised Feature Selection

Abstract

Objectives: To propose hybrid feature selection technique based on Particle Swarm Optimization and Genetic Algorithm for phonocardiography. Methods/Statistical Analysis: The system estimated using heart sounds corresponding to different heart conditions like 320 signals out of which 150-normal, 70-Mitral Valve Prolapse, 50-Ventricular Septal Defect and 50-Pulmonary Stenosis. Features are extracted using DWT. Finding: A phonocardiographic signal reflects the health status of the heart. Generally there exists two heart sounds, but further sounds indicate disease. Phonocardiography is non-invasive, low-cost and accurate method to detect heart disease. This work proposes a framework to extract information from phonocardiography signal to classify whether it is proper or improper. Discrete Wavelet Transform method is implemented to extract features followed by Singular value decomposition for feature selection. Applications/Improvements: pplications/Improvements: An experimental result gives the improvement of the proposed method by increasing the efficacy of the corresponding feature selection technique.