Heart Sound Classification Using Weka for Machine Learning

The accurate and efficient classification of heart sounds represents a crucial challenge in cardiac care, with significant implications for early diagnosis and timely intervention. This study investigates the application of machine learning techniques, utilizing the Weka framework, to address this problem and used PhysioNet dataset, which comprises 3,240 heart sound recordings, and extracted 238 features to capture the essential spectral and temporal characteristics of heartbeats. Considering the high dimensionality of the dataset, various feature selection methods were evaluated, including information gain ranking, Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE). Ultimately, the InfoGain algorithm was applied, selecting the 50 most significant features from the initial set of 238. The performance of five machine learning models—Multilayer Perceptron, Random Forest, Decision Tree, k-Nearest Neighbor (k-NN), and Support Vector Machine (SVM) — on both the full and reduced feature sets, using 80-20 split and 10-fold cross-validation was evaluated. The results demonstrate that the Random Forest model achieved the highest accuracy of 92.1% on the full feature set, while the k-NN model exhibited superior computational efficiency on the reduced set, with an accuracy of 89.4% and significantly faster training times. These findings highlight the trade-offs between model complexity and efficiency, providing valuable insights for clinicians and researchers in selecting appropriate algorithms for real-world heart sound classification applications.