Detection of Arrhythmia via Electrical Activity of the Heart Using AI Techniques

Abstract

Electrocardiogram (ECG) consists of several waveforms such as the P wave, QRS complex, and T-wave. Various heart diseases can be diagnosed by observing the variations in the shape of each waveform and the distance between different peaks. There are 15 suggested classes for arrhythmia which are divided into 5 super classes: Normal (N), Supraventricular ectopic beat (SVEB), Ventricular ectopic beat (VEB), Fusion beat (F), and Unknown beat (Q). The input signal is classified into these 5 categories using 1D Convolution Neural Network, Random Forrest Classifier, Decision Tree Classifier, and MPL Classifier. In the clinical diagnosis of cardiac illness, the ECG classification is crucial. The three main performance metrics—precision, recall, and F1 score are calculated for all 4 models and compared. The accuracy of the four models is also computed and compared. The One-Dimensional Convolution Neural Network achieved 98% accuracy, 90% macro average, and 98% weighted average. The Decision Tree Classifier achieved 95% accuracy, 80% macro average, and 95% weighted average. The Random Forest Classifier achieved 97% accuracy, 86% macro average, and 97% weighted average. The MLP Classifier achieved 98% accuracy, 87% macro average, and 97% weighted average.