Research on Heart Sound Classification Based on SVM Model

Heart sound is an important physiological signal that is of great significance for evaluating heart function and discovering heart diseases. This article selects 610 sets of heart sound audio from the publicly available datasets of PhysioNet/Cinc Challenge 2016 and PASCAL Challenge. Through wavelet transform denoising and Shannon theory extraction of envelopes, binary processing is performed to obtain key information of heart sounds and extract six characteristic indicators of heart sounds. Finally, these indicators were used as input parameters for machine learning, and SVM algorithm was used to classify heart sound signals, resulting in a total of ten cases with recognition accuracy of over 81.67%. The accuracy, recall, and F1 score of the SVM model are all above 0.7. The results indicate that wavelet transform and Shannon envelope extraction can ensure the quality of heart sound signals and the acquisition of effective features. The SVM model can effectively classify heart sound signals, providing a way to assist clinical decision-making.