Machine Learning Analysis of Heartbeat Sounds Using Weka for Classification

Classifying heartbeat sounds is a critical step in improving cardiovascular diagnostics, especially as reliance on traditional stethoscope assessments can be limited by human subjectivity. In this study, we explored the application of machine learning techniques to the PhysioNet Heartbeat Sound Database, aiming to differentiate normal from abnormal heart sounds with both accuracy and consistency. Our approach integrates unsupervised K-means clustering, which identifies natural groupings within the data, with supervised classification models to validate these groupings and enhance predictive power. Key acoustic features-such as pitch, RMS energy, MFCCs, and formants-were extracted to provide a comprehensive representation of the heart sounds. To streamline classification, we applied the InfoGain algorithm, selecting the 50 most significant features from an initial set of 238. This process ensured that the models focused on the most relevant information, improving efficiency and interpretability. We evaluated four classifiers-SVM (SMO), Decision Tree (J48), Naive Bayes, and k-NN (IBK)-using a 10-fold cross-validation approach. SMO and J48 demonstrated strong performance, achieving accuracies of 97.4% and 96.02%, respectively, with high Kappa statistics, showcasing their reliability in distinguishing heart sound categories. These results highlight the potential of machine learning in addressing critical healthcare challenges, particularly when labeled data is scarce. Future work will focus on expanding the dataset, incorporating neural network architectures, and exploring additional clustering methods to further improve generalizability and diagnostic accuracy.