Heart Sound Classification with MFCCs and Wavelet Daubechies Analysis Using Machine Learning Algorithms.

Area of Science:

• Biomedical Engineering
• Cardiology
• Machine Learning

Background:

• Cardiovascular diseases (CVDs) are a leading global cause of mortality.
• Accessible early detection tools are crucial for managing CVDs.
• Automated classification systems offer non-invasive diagnostic support.

Purpose of the Study:

• To implement and evaluate machine learning models for classifying normal and abnormal heart sounds.
• To assess a hybrid feature extraction approach combining MFCCs and wavelet analysis.
• To develop a computationally efficient tool for computer-aided auscultation.

Main Methods:

• Utilized the PASCAL dataset of heart sound recordings (normal, murmur, extrasystolic).
• Extracted statistical features using Mel-Frequency Cepstral Coefficients (MFCCs) and Daubechies wavelet analysis.
• Trained and compared four classifiers: SVM, logistic regression, random forests, and decision trees.

Main Results:

• All classifiers demonstrated notable performance in distinguishing heart sound classes.
• The Support Vector Machine (SVM) model using 26 MFCCs and Daubechies-4 wavelet coefficients achieved the highest accuracy.
• Performance was evaluated using accuracy, precision, recall, specificity, F1-score, and AUC.

Conclusions:

• The hybrid MFCC-Wavelet framework offers competitive diagnostic accuracy for heart sound classification.
• This approach provides a lightweight, interpretable, and computationally efficient solution.
• The findings support the use of this framework for early cardiovascular screening and computer-aided diagnosis.