An Ensemble Convolutional Neural Network-Bidirectional Long Short-Term Memory-Attention Approach for Electrocardiogram-Based Heart Disease Diagnosis

The global prevalence of cardiovascular diseases demands urgent and precise electrocardiogram (ECG) analysis for proper diagnosis because these diseases lead to numerous deaths each year. The proposed research presents an ensemble deep learning framework which utilizes convolutional neural networks (CNNs) with bidirectional long short-term memory (BiLSTM) layers combined with attention mechanisms to identify temporal and spatial patterns in raw ECG recordings. The preprocessing step includes normalization followed by overlapping segmentation of multi-channel ECG signals to extract dynamic signal patterns. Individual ensemble members use the combination of two-stage CNN architecture with residual components for local feature extraction followed by BiLSTM component that models temporal dependencies and an attention layer that enhances discriminative sequence parts before fully connected layers produce the final classificationThe ensemble approach demonstrates strong performance on the PTB Diagnostic ECG Database by reaching high accuracy coupled with precision and recall and F1 scores as well as superior results than single-model architectures. Through deep feature extraction with CNNs with sequential modeling using BiLSTM along with an attention mechanism for providing salient feature weights the approach provides better generalization and reduced overfitting. The framework unites various modeling methods to establish a possible real-time ECG-based diagnostic system which can improve reliable cardiovascular disease screening in medical environments.