Adventitious and Normal Respiratory Sound Analysis with Machine Learning Methods

Abstract

The computerized respiratory sound analysis systems provide vital information concerning the current condition of the lung. These systems, used by physicians for the diagnosis of diseases, help to classify respiratory sounds. Because each physician has different knowledge and experience, there is a problem with diagnosing and treating respiratory system diseases. This study will help the physician to decide in various difficult diagnostic situations easily. For this purpose, different machine learning classifiers and feature extraction models have been constituted to classify respiratory sounds as healthy and patient then its results were compared. In this study, Empirical Mode Decomposition, Mel Frequency Cepstral Coefficients, and Wavelet Transform methods are used for feature extraction, while k Nearest Neighbor, Artificial Neural Networks, and Support Vector Machines are used for classification. The best accuracy was 98.8% by using combination Mel Frequency Cepstral Coefficient and k Nearest Neighbor methods.