An Automatic ECG Classification for Arrhythmia Detection Using Machine Learning Techniques

Cardiovascular arrhythmias significantly impact global health, with the electrocardiogram (ECG) serving as a primary diagnostic tool due to its accessibility and cost-effectiveness. Despite its widespread use, ECG misdiagnosis rates remain high, often requiring extended patient monitoring to detect infrequent arrhythmias. Manual diagnosis is time-consuming and heavily reliant on cardiologist expertise. Consequently, automatic arrhythmia detection techniques have been developed worldwide. However, many machine learning-based ECG classification systems face challenges in generalization and transparency, essential for clinical trust and verification. This study presents an automated approach for detecting and classifying eight types of arrhythmias using the MIT-BIH dataset. Two validation protocols, holdout beats and leave-out patients, were employed for data splitting and resampling. The performance of Support Vector Machine (SVM) and MultiLayer Perceptron (MLP) classifiers was compared against other state-of-the-art classifiers. The proposed models demonstrate strong performance, particularly under the leave-out patients' validation protocol, indicating superior generalization capabilities. This research underscores the potential of MLP and SVM in enhancing the accuracy and reliability of arrhythmia detection and classification, thereby improving diagnostic outcomes and patient care.