Classification of Myocardial Infarction using MFCC and Ensemble Subspace KNN

Myocardial Infarction and Heart stroke are most prominent in the top ten deadly diseases worldwide. These diseases remained the prime reason for mortality in the last fifteen years. There is a need to develop a system capable enough to detect the disorders efficiently in a reliable manner. We have proposed an algorithm by using Phonocardiography (PCG) to classify different categories of Myocardial Infarction (MI). The database used in this study consists of self-collected PCGs acquired from various hospitals. Mel-frequency Cepstral Coefficients (MFCCs) are extracted directly from the acquired data of heart sound for key feature extraction and performed classification by Ensemble Subspace K Nearest Neighbor (KNN) method in MATLAB 2019a. Simulation results evidenced that the proposed algorithm achieved a mean accuracy of 94.9%. Further research can be done to design an embedded system of the proposed methodology to assist medical specialists in clinical diagnosis.