ECG Classification with Dual Models: XGBoost Voting and Deep Learning with Attention

During the process of Electrocardiogram (ECG) recognition, frequent challenges arise concerning the lack of convenience and accuracy. Moreover, the imbalanced distribution of ECG data presents significant hurdles for ECG recognition models. To address the aforementioned issues, the paper proposes two ECG recognition models on a dataset enhanced through Generative Adversarial Network (GAN): XGVoting and CRANet, which are specifically designed to tackle the challenges of lightweight design and accuracy improvement. By utilizing GAN for data augmentation, the ECG dataset is effectively balanced. The first model, XGVoting, is a lightweight approach that intelligently combines the ensemble tool eXtreme Gradient Boosting (XGBoost) with voting, thereby enabling swift and efficient detection. The second model, CRANet, is meticulously designed to achieve high-accuracy recognition by effectively integrating Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and self-attention mechanisms. The efficacy of the proposed models is validated using the Massachusetts Institute of Technology-Beth Israel Hospital Arrhythmia Database (MIT-BIH). Remarkably, under the premise of lightweight design, XGVoting attains an impressive overall accuracy of 97.37%, while ensuring a perfect 100% accuracy in identifying normal heart-beats. Meanwhile, CRANet exhibits outstanding performance with an overall accuracy of 99.0%.