Computer-aided Phonocardiogram Classification using Multidomain Time and Frequency Features

This paper presents an improved classification technique for automated classification of phonocardiogram (PCG) signals. In the light of presented literature study, a number of representative multidomain time and frequency features are suggested for the heart signal analysis and classification with comparatively large and imbalanced dataset. Machine learning algorithms such as support vector machines (SVM), k-nearest neighbor (KNN), Decision Tree (DT) and TreeBagger (TB) are tested for heart sound (HS) classification. For the performance evaluation metrics such as, accuracy, final score, sensitivity and specificity are computed for each classifier. Overall, all the classification algorithms performed well by achieving final scores greater than 85% but with the designed setup and dataset SVM outperformed others by achieving final score of 94.20% (Accuracy 95.31%, Sensitivity 92.30%, Specificity 96.08%).