ECG Heartbeat Classification Using Ensemble of Efficient Machine Learning Approaches

Electrocardiogram, an established method for cardiac health analysis, has attracted significant research interest in accurate heartbeat classification. Despite numerous studies in this field, achieving high accuracy scores remains a challenge. This paper employs and fine-tunes well-known machine learning techniques, comparing them with additional cutting-edge techniques. The study utilizes highly imbalanced datasets, addressing this issue by adjusting the Artificial Neural Network (ANN) and Residual Network (ResNet) models loss value through class weight assignment. Two enriched ECG datasets are used: the PTB Diagnostic ECG dataset, which has two classes, and the MIT-BIH Arrhythmia dataset, which has five classes. For the MIT-BIH Arrhythmia dataset, the suggested methods attain accuracies of roughly 98% (ANN) and 95% (ResNet); for the PTB Diagnostic ECG dataset, they obtain accuracies of 97% (ANN) and 94% (ResNet). The results of this study outperform those of other cutting-edge approaches in both cases.