Automatic classification of respiratory sounds with resampling and machine learning techniques

Respiratory signals emitted from human lungs provide vital and decisive information about the health status of a patient's lungs. Traditional clinical methods require professional pulmonologists to accurately diagnose such signals. Computer aided identification systems are frequently used in this field. However, sufficient data is critical to the high performance of Computer Aided Identification systems. In this proposed study, an automatic diagnostic system for normal/abnormal classification from resampled respiratory sound data is proposed. First, more than 200 sound features were extracted from the resampled sound data, and then the feature was selected and presented to the prediction model. Traditional Machine Learning models such as k-Nearest Neighbor, Support Vector Machine, Naive Bayes, Decision Tree and Random Forest are used for prediction models. For each classification method, 10-fold Cross Validation was performed and results between 93% and 95% were obtained.