AN INNOVATIVE MACHINE LEARNING FRAMEWORK FOR PHONOCARDIOGRAPHY (PCG) USING MFCC AND DEEP EXTREME LEARNING MACHINE (DELM)

Cardiovascular Diseases (CVDs) are a significant global cause of mortality, necessitating effective diagnostic techniques. Phonocardiography (PCG) is among the fundamental methods used to analyze heart sounds to detect human heart-related abnormalities. However, in an environment where state-of-the-art PCG equipment is not available, a Machine Learning (ML) based solution can serve as a reliable alternative. However, the main challenges faced by ML-based PCG systems, are the unavailability of balanced and unbiased datasets, the vanishing and exploding gradient a well-known Deep Learning (DL) issue, and inappropriate feature extraction techniques, which often compromise the accuracy and reliability of ML-based PCG systems. This study introduces a novel Deep Extreme Learning Machine (DELM) and Mel-Frequency Cepstral Coefficients (MFCC) based PCG framework for CVD diagnosis. The proposed framework uniquely addresses the above mentioned challenges. The proposed model achieves a remarkable training accuracy of 98.46 % and a test accuracy of 86.80 %, using the Heartbeat Sound dataset with five classes and after class aggregation and dataset normalization the proposed model achieved training accuracy 99.52 % and a test accuracy of 92.30 % demonstrating its potential in PCG diagnostics. This framework represents a significant advancement in ML-based PCG systems for automating heart sound analysis and contributing to improved cardiac healthcare, especially in resource-limited settings.