Classification and Detection of Heart Rhythm Irregularities using Machine Learning

Accurate detection of arrhythmias is critical to patient well-being in clinical settings because the readings usually reflect acute and chronic cardiac problems. Because of variability between individuals and unavoidable noise, this is known to be a difficult problem even for human experts. This paper explores the use of deep neural networks for the task of classifying ECG recordings using recurrent and residual architectures. The proposed method tested on ECG dataset of 162 patients' readings that consist of three classes including normal sinus rhythm, cardiac arrhythmia and congestive heart failure. Results shows that proposed LSTM method having 4 hidden layers and optimization function as Adam gives maximum accuracy of 99.12% in comparison to other methods.