Cardiovascular disease (CVD) has been the most common factor of death for decades, and one method to detect CVD is through heart sound auscultation. Numerous studies have investigated improvements in precision and accuracy for heart sound classification using machine learning. Nonetheless, most methods utilize many features in their machine learning to increase the accuracy of their predictive model to address challenges associated with signals acquired through sensors placed at different locations. In this paper, we propose the use of heart sounds segmented into three frequency bands and the extraction of features, namely, the Shannon entropy and spectral entropy of each frequency band, to serve as an input to our support vector machine (SVM). The focus of the study is to examine the use of only six features to achieve a satisfactory score in heart sound classification. The technique is assessed using an online heart sound database. The features that were extracted are trained and tested using the SVM to predict normal and abnormal heart sounds. Results demonstrated accuracies of 95 and 78% for normal and abnormal heart sounds, respectively. Subsequently, the testing results achieved an overall accuracy of 82.5% with a sensitivity of 85% and a specificity of 80%.