Detection of Lung Diseases from Acoustic Recordings using Statistical Moments of MFCC and Machine Learning

Lung diseases are the third leading cause of mortality globally, often remaining undetected until advanced stages. This study aims to accurately detect lung diseases using acoustic recordings by using clinically correlated features and machine learning models. Initially, audio signals are preprocessed with a high-pass filter to eliminate noise and artifacts and then Mel-Frequency Cepstral Coefficients (MFCCs) are employed. Further, key statistical features are extracted from MFC coefficients using Mean, Standard deviation, and Skewness. The different subsets of these features are used to train the machine learning models. The detection performances of these features are evaluated on both binary and multiclass lung diseases using k-Nearest Neighbour (k-NN), support vector machines (SVM), and random forest models. The results demonstrate significantly encouraging detection performances, highlighting the potential of these features and machine learning approaches in accurately diagnosing a range of lung diseases.