Comparative Performance Analysis of Heart Sound based Heart Disease Classifiers Using Machine Learning Models from Balanced Datasets

Accurate heart disease (HD) detection and follow up treatment at an early stage can save lot of human lives. In recent years, many research efforts have been made to achieve this objective. The HD detection using machine learning (ML) approach can be made either from various test reports, electrocardiogram (ECG) signal or heart murmurs. In recent past many ML based classification models have been reported for HD detection from heart sound signal. But still there is further scope to develop improved and reliable intelligence models for HD detection. In this paper three popular and well-performing ML techniques such as Random Forest (RF), K-nearest Neighbour (KNN) and Extreme Gradient Boosting (XGB) have been employed for classification of HD using heart sound signal from two standard data bases. The features extracted from Discrete Wavelet Transform (DWT), Mel Frequency Cepstral Coefficients (MFCC) and the combination of the two features are used as inputs to the models for development and testing purposes. Analysis of the results obtained from the simulation study reveals that the RF model outperforms the other two models in terms of accuracy, F1-score and Area Under the Curve (AUC). Further, it is observed that the combined features-based RF classifier performs the best. The proposed model can also be employed for the detection of other diseases after suitable training.