ECG signal classification via ensemble learning: addressing intra and inter-patient variations

Electrocardiogram (ECG) signal classification is a cornerstone of automated heart abnormality detection. Unlike the limitations of human interpretation, AI techniques can effectively identify subtle patterns in ECG signals. This makes ECG a powerful non-invasive tool for assessing cardiovascular health. Existing methods for classifying ECG signals while valuable, they still struggle to achieve both high sensitivity and specificity. This limitation hinders their ability to deliver accurate and timely diagnoses for cardiac conditions. These shortcomings emphasize the need for more effective techniques to improve the precision of ECG signal classification. In response to these challenges, this study introduces a novel approach, using an ensemble methodology, a machine learning technique to enhance the precision of ECG classification through the fusion of signal and wave features. The proposed methodology addresses two key challenges: the transformation of paper ECG recordings into one-dimensional digital signals amenable to machine learning algorithms and the automated extraction of diagnostically significant features including the P wave, QRS complex, and T wave. Validation of the proposed methodology encompasses a comprehensive evaluation on a heterogeneous dataset comprising real-world and publicly available online resources. Noteworthy aspects of the evaluation include considerations of both intra-patient variations and inter-patient discrepancies, thus reflecting real-world complexities. Notably, in the realm of machine learning, the study employs ensemble algorithms and a soft voting classifier to enhance classification accuracy and robustness. This paper contributes to the advancement of automated ECG classification, offering a promising avenue for precise and reliable cardiovascular health assessment.