Machine Learning Based Automatic Classification of Respiratory Signals using Wavelet Transform

Respiratory signals emanating from human lungs give vital and indicative information regarding the health status of a patient's lungs. Conventional clinical methods require professional pulmonologists to diagnose such signals properly and are also time consuming. In this proposed work, an efficient and automated method is proposed for the diagnosis and classification of respiratory signals into normal and abnormal respiratory sound. Respiratory signal is cleaned using a band pass filter, followed by features extraction in wavelet domain. Discriminatory features from the filtered signals are fed to SVM for purpose of classification of signals. Proposed methodology has achieved an accuracy of 92.30% in correctly classifying the pathological lung sounds. Outcomes of the proposed algorithm are promising and indicates its usability for some real time application.