Classification of Normal /Abnormal Heart Sound Recordings

Abstract

In this project, we address the medical field related to the cardiovascular system and heart diseases. Our objective is to develop a solution based on artificial intelligence techniques, in particular those of Machine Learning, in the form of an intelligent diagnostic support system to detect heartbeat anomalies using the dataset of the 2016 PhysioNet/CinC Challenge. For this we apply techniques for extracting temporal characteristics from PCG signals. In order to train the model to recognize heart sounds, we used various algorithms such as Random Forest, KNN and SVM. The results obtained are very satisfactory with an accuracy score of 92% and demonstrate the effectiveness and reliability of our intelligent system, which aims to be a diagnostic aid tool for health practitioners.