Multi-Class Arrhythmia Classification from ECG Signals Using Deep Learning

Abstract

Arrhythmias pose a serious health risk, and healthcare professionals require reliable, immediate methods to detect irregular heartbeats. This study evaluates the capacity of machine learning models—specifically Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and k-Nearest Neighbors (k-NN)—to recognize multiple arrhythmia types from electrocardiogram (ECG) signals. The hierarchical feature extraction ability of CNNs makes them effective at capturing morphological ECG structure, while RNNs, particularly LSTM and GRU variants, excel at modelling the temporal dependencies present in ECG time series. The traditional k-NN algorithm classifies heartbeats by comparing heart-rate and PQRST-wave features. Public ECG databases are used to train and evaluate the surveyed models, and performance is reported in terms of accuracy, sensitivity, and specificity. This review synthesises the comparative strengths of deep learning and traditional machine learning approaches for multi-class arrhythmia detection and proposes a hybrid CNN-LSTM-Attention framework together with a detailed experimental protocol for future cardiac diagnostic technology development.