The increasing prevalence of cardiovascular diseases (CVDs) has become a major health concern. Arrhythmia is the deadliest heart condition of all cardiovascular disorders. Thus, timely and precise arrhythmia diagnosis is critical in preventing heart disease and abrupt cardiac death. Arrhythmia can be discovered on an electrocardiogram (ECG) by observing irregular heart electrical activity. The heart's electrical activity is recorded as an ECG signal, which contains both normal and pathological information. Classification of ECG patterns is critical for automatically diagnosing cardiac illness. This paper discusses the various learning approaches for automatically distinguishing different types of heartbeats. According to reported studies, the convolutional neural network (CNN) model is the best option for classifying arrhythmia. An ensemble of depth wise separable convolutional (DSC) neural networks achieves the highest classification accuracy, 99.88%.