ECG-based Heart Arrhythmia Recognition using Enhanced Extreme Learning Machine

Abstract

Reliable classification of electrocardiogram (ECG) signal patterns is fundamental for early diagnosis and effective clinical evaluation of cardiovascular abnormalities. This paper proposes a hybrid system for accurate classification of cardiac arrhythmias using ECG signals. A convolutional neural network (CNN) extracts complex features, and an Enhanced Extreme Learning Machine with Leave-One-Out (LOO) cross-validation classifier then provides accurate and stable models during training. A data augmentation module balances the five AAMI heartbeat classes of the publicly available MIT–BIH Arrhythmia Database. In tests on this database, the CNN-EELM-LOO model reaches an accuracy of 99.81% (intra-patient evaluation) and improves on classical machine learning and recent deep learning methods. The Leave-One-Out cross-validation integrated into the decision model increases overall robustness and reduces the effect of random initialization and the bias that can remain even with repeated training. This research demonstrates that the hybrid EELM-LOO architecture is an effective tool for automated arrhythmia detection, with strong potential for integration into clinical decision support systems.