Comparative Analysis of Deep Learning and Traditional Machine Learning Models for Arrhythmia Classification using ECG Signals

Abstract

Arrhythmias, a form of cardiovascular disease, are a major contributor to the high global mortality rate. Early detection of arrhythmias through electrocardiogram (ECG) analysis can significantly improve patient outcomes. This study investigates the application of var- ious machine learning (ML) and deep learning models for the classification of arrhythmias using ECG signals from the MIT-BIH Arrhythmia Dataset. The models evaluated include Random Forest, Support Vector Machines (SVM), Logistic Regression, Multilayer Perceptron (MLP), and Convolutional Neural Networks (CNN). Additionally, feature selection techniques, such as the Fourier Transform, were applied to enhance the performance of the ML models. Among the models tested, the CNN achieved the highest accuracy (89.29%), F1 score (85.69%), and AUC (87.98%), demonstrating its superior ability in accurately detecting arrhythmias. In contrast, traditional ML models, including Random Forest and SVM, showed moderate performance with lower accuracy and discriminatory power. The study highlights the potential of CNN-based architectures for automated ECG analysis and emphasizes the importance of integrating explainable AI techniques to increase the transparency and clinical adoption of deep learning models. Future research could focus on larger, more diverse datasets and the use of Recurrent Neural Networks (RNNs) for longer ECG recordings to improve classification performance further.