Incorporating MFCC Feature Extraction to the Classification of Respiratory Sounds by Machine Learning Algorithms

Respiratory diseases such as asthma, chronic obstructive pulmonary disease (COPD), and lower respiratory infections are among the leading causes of death globally, particularly in remote areas. Disease diagnosis is often performed using traditional methods such as auscultation, which relies on the subjective expertise of medical professionals. As machine learning algorithms gain prominence in medical data analysis, they provide a more accurate alternative for analyzing respiratory sounds. The purpose of the research is to compare the performance of several machine learning models in classifying respiratory sounds. The method used is a machine learning-based approach utilizing Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction, followed by classification algorithms such as Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), Naïve Bayes and Deep Neural Networks (DNN). The result of this study shows that the Random Forest classifier exhibited a superior accuracy of 79%. In contrast, SVM, DNN, and KNN classifiers demonstrated stable accuracies ranging from 75% to 77%, and Naïve Bayes recorded the lowest of 52%. Across all accuracy metrics, the Random Forest outperformed the rest of the models as the most effective choice for classifying the respiratory sound. This research provides a foundation for developing more accurate diagnostic tools for respiratory abnormality.