HEART SOUNDS CLASSIFICATION WITH DEEP FEATURES AND SUPPORT VECTOR MACHINES

In this study, the heart which is one of the most important organ affecting life-sustaining function is examined whether it works properly in a certain rhythm or not. In this regard, an effective algorithm both analyzing and categorizing Phono-cardiogram signals (PCG) which is significant at diagnosis of diseases is presented. First of all in this context, as colored spectrogram images of heart sounds are established to be able to analyze PCG signals, the characteristic extraction maps of Convolutional Neural Networks (CNN) are used to educate the data of images obtained. CNN-VGG16 model educated previously is used when these maps are established and and it is categorized with Support Vector Machine (SVM) which is an effective classifier at machine education. The performance of all rating labels is evaluated separately for experimental study with two different data. While max performance improves about %8 in one data set (DATASETA), max performance is obtained at other dataset (DATASETB) for normal rating label.