Respiratory sound classification : different respiratory sounds

• Main Author: Oon, Shawn Guowei
• Other Authors: Ser Wee
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/150734
• Institution: Nanyang Technological University

Chronic Obstructive Pulmonary (COPD) Disease is commonly misdiagnosed until it is too late. The primary cause of COPD is smoking, air pollution. Their symptoms are breathlessness, chronic coughing, mucous production. As the symptoms worsen, it would result in the need of urgent medical care or even death The aim of this project is to study and analysis the Healthy, Crackles and Wheezing and find out the accuracy of obtaining the right prediction in the machine learning comparison of 2 classes. Audio samples were collected from an open-source database, ICBHI 2017 Challenge, and some from TTSH. Audio samples go through a series of python algorithm and come out with a predicted outcome accuracy. Using the mean and variance values obtained from the MFCC, Fisher’s Ratio is used to calculate the discriminant of the feature. With the comparison of Crackles & Healthy, features 3,4,5,7 and 13 is the top 5 features. The top 5 features for Crackles & Wheezing are 2,4,7,8 and 9. For Wheezing & healthy, the features 3,7,8,9 and 13. Confusion Matrix is being used to determine the how good the classifier is and whether the accuracy obtained is reliable. Both Wheezing & Healthy and Crackles & Healthy has obtained good accuracy, precision, recall and F1-score for both train and test set. But for Crackles & Wheezing, only the training set produces good result and not for the test set. Hence it shows that it may not be a good indicator when testing. From the analysis of K-Fold Cross Validation, it is shown that Healthy and Wheezing, Healthy and Crackles can produce reliable test accuracy. While Wheezing and Crackles may not be a good indicator for testing as their accuracy for the confusion matrix is 64%.