Implement and Accelerate a Deep RNN with BiLSTM based ECG Signal Classification for Heart Disease prediction in Internet of Things Network

Abstract

Modern healthcare systems are increasingly challenged by the growing prevalence of cardiovascular diseases, necessitating accurate and automated ECG-based diagnostic solutions. In order to effectively classify ECG signals into normal and abnormal categories, this study suggests a hybrid deep learning framework based on Recurrent Neural Networks (RNN) and Bidirectional Long Short-Term Memory (BiLSTM). To guarantee robustness and generalization across several data splits, the suggested model is assessed using a K-Fold cross-validation technique on the PTB-XL dataset. A standardized preprocessing pipeline is applied, including signal normalization and appropriate label transformation, while class imbalance is handled through controlled data distribution strategies during training to maintain evaluation integrity. The hybrid architecture leverages temporal sequence learning & bidirectional context capturing, enhancing feature representation and classification performance. Experimental results at Fold 5 demonstrate strong & consistent performance, achieving up to 95.32% accuracy, 99.16% precision, 96.08% recall, 97.60% F1-score & a ROC-AUC of 0.70 across folds. The high precision indicates reliable abnormal detection, while variation in specificity reflects inherent dataset imbalance challenges. Comparative analysis across folds confirms the stability and effectiveness of the proposed model, showing improved performance over conventional single DL approaches. The results highlight the capability of the hybrid Deep RNN-BiLSTM framework for accurate ECG classification, making it a promising solution for real-time clinical decision support systems.