Advanced ECG classification with improved optimized feature selection and attention CNN-BILSTM technique

Accurate prediction and classification of ECG signals are crucial for effective cardiac disease diagnosis due to their role in early detection, personalized treatment, and ongoing disease management. Precise classification allows for the early identification of cardiac abnormalities, enabling timely interventions that significantly improve patient outcomes and prevent severe complications. Traditional methods often struggle with the dynamic nature of ECG signal data, leading to suboptimal performance in identifying abnormalities. To mitigate these shortcomings, this research introduces a novel approach for ECG signal classification. The proposed framework begins with normalization of the ECG signals to reduce noise and maintain uniformity throughout the dataset. Following normalization, feature extraction is performed using an Improved Wavelet Transform (IWT), offering enhanced capabilities for capturing the intricate details and temporal changes in ECG signals by transforming them into different frequency components. To further refine the feature set, the Improved Optimized Bird Swarm Algorithm (IOBSA) is employed, which efficiently identifies the most relevant and discriminative features while reducing dimensionality. Finally the selected features are processed by an Attention Convolutional Neural Network (CNN) combined with Bidirectional Long Short-Term Memory (BiLSTM) networks for classification. The Attention CNN enhances the model’s ability in focussing on critical regions of ECG signals, while the BiLSTM component captures both past and future dependencies in data, improving the overall classification performance. This integrated framework is executed using Python software, and results shows significant improvements in classification accuracy compared to conventional approaches.