Classification of Phonocardiography Signals Using Imbalanced Machine Learning Techniques

Cardiovascular diseases, which involve heart and blood vessel dysfunctions, cause a higher number of deaths than any other disease in the world. Throughout history, many approaches have been developed to analyze cardiovascular health by diagnosing such conditions. One of the methodologies is recording and analyzing heart sounds to distinguish normal and abnormal functioning of the heart, which is called Phonocardiography. With the emergence of machine learning applications in healthcare, this process can be automated via the extraction of various features from phonocardiography signals and performing classification with several machine learning algorithms. Many studies have been conducted to extract time and frequency domain features from the phonocardiography signals by segmenting them first into individual heart cycles, and then by classifying them using different machine learning and deep learning approaches. In this study, various time and frequency domain features have been extracted using the complete signal rather than just segments of it. Random Forest algorithm was found to be the most successful algorithm in terms of accuracy as well as recall and precision.