Heart Beat Sound Signal Classification Using 12-Layer Neural Networks

Heart diseases, including coronary artery disease, irregular heartbeats, congenital heart defects, and heart valve disease, are often caused by unhealthy lifestyle choices. Unfortunately, the treatments for these diseases can involve surgeries, transplantation, and heavy doses of medication. The objective of this paper is to classify a person's heartbeat as regular or irregular using descriptive statistics in advance by identifying the variation in S3 heart sound in abnormal heart sound signal. This approach enhances simplicity and is comprehensible for everyone since descriptive statistics summarizes a dataset using three fundamental classes of measures: central tendency, variability, and frequency distribution. The artificial neural networks replicate the human brain with the help of input, hidden and output layers. By analyzing and extracting descriptive statistics of the normal and abnormal heartbeat sound signals whose datasets are balanced using SMOTE technology, an artificial neural network model is built with one input, ten hidden, and one output layer as an architecture. The model is then trained and optimized with a hundred epochs to achieve high accuracy of 88% and minimum loss.