Machine Learning Based Lung Sound Analysis Using MFCC Features

Lung sound classification plays a major role in detecting abnormalities such as wheezing, crackles, or diminished breath sounds, which may indicate the presence of lung diseases like asthma, Chronic Obstructive Pulmonary Disease (COPD), pneumonia, or pulmonary fibrosis. In this paper, we propose an automated machine learning algorithm for the classification of lung sounds using Mel-Frequency Cepstral Coefficients (MFCC) features. The proposed method is tested using the ICBHI (International Conference on Biomedical and Health Informatics) challenge dataset. Based on benchmark performance metrics, the proposed method achieved an accuracy, sensitivity, specificity, and precision of 78%, 86%, 64%, and 77%, respectively, in classifying lung sounds as normal or abnormal (wheezes or crackles, crackles and wheezes) lung sounds. Performance results showed that a processing time of 0.29 seconds is required to classify the lung sound as normal or abnormal. The proposed method has great potential to assist in the early identification of respiratory problems, allowing for timely intervention and management.