A Review of Machine Learning Methods for Screening Cardiac Abnormalities using Auscultation

This paper presents a thorough review of various methodologies employed in heart sound classification, combining both conventional machine learning (ML) algorithms and complex deep learning (DL) approaches. We systematically evaluate different data collection approaches, preprocessing techniques, methodologies, and classification mechanisms. Further-more, the review analyzes several feature extraction methods required for capturing relevant signal characteristics, spanning Mel-frequency cepstral coefficients (MFCC), continuous wavelet transform (CWT), and convolutional neural networks (CNNs). Additionally, we examine a variety of classification methods, from traditional methods like support vector machines (SVMs), Knearest neighbors (KNNs), and Decision Trees (DTs), to more sophisticated parametric approaches such as recurrent neural networks (RNNs), (CNNs), long-short term memory (LSTM), and even ensemble techniques, signifying the potential of combining classification approaches. Key findings reveal a trend toward employing hybrid models that merge the capabilities of multiple techniques to enhance cardiac diagnostics precision. This comprehensive overview aims to explore existing methodologies and move toward bridging the gap in today's technologies, establishing a robust diagnostic tool for medical applications.