Diagnosis of Valvular Heart Disease Based on Ensemble Learning

Heart sound signal processing consists of different phases. After applying necessary preprocessing and segmenting heart sound cycles, some distinctive features of heart sound are extracted. Since the appropriate operation of the classifier has a high impact on the performance of the system, in this study we propose a proper classification algorithm. One of the commonly used methods to build accurate classifiers is to use a group of classifiers and make decision based on the outputs of these classifiers. By far, the performance of the ensemble methods has been investigated in different fields of classification problems by researchers. However, in the field of heart valve diagnosis there are almost no studies investigating these methods. In this study, we train several linear classifiers and the final decision is made according to the outputs of them based on the majority voting algorithm. The training samples of each classifier are chosen randomly with replacement from the whole training set. The proposed method is implemented for 5 datasets and also compared with 3 other methods using different criteria including sensitivity, specificity, diagnostic odds ratio, precision and error. Results show that the proposed method has higher accuracy and faster prediction time. The noise label problem and the robustness of the proposed method against this noise are also investigated. Statistical tests show that the proposed method significantly outperforms other methods.