A Hybrid CNN-LSTM Framework for ECG Classification with Genetic Algorithm-Based Feature Optimization

Objective: The aim is to improve the accuracy and effectiveness of ECG signal classification, which is crucial for the premature detection of cardiovascular diseases. Our primary aim was to develop a robust, lightweight hybrid model that achieves a high classification rate while maintaining low computational complexity, making it suitable for real-time clinical applications. Methods: A deep learning hybrid model, consisting of a Convolutional Neural Network (CNN) with Long Short–Term Memory (LSTM), is introduced. For feature selection, a Genetic Algorithm (GA) is employed, utilizing a Support Vector Machine (SVM) as the fitness function of the GA. Principal Component Analysis (PCA) is used for dimensionality reduction. Denoising, normalization, and segmentation of ECG signals from the PTB-XL dataset. CNN layers detect spatial features, while LSTM layers capture temporal relationships. The Adam optimizer is used to enhance training convergence. Findings: The proposed CNN-LSTM + GA-SVM-PCA model achieved precision, classification accuracy, recall, specificity, and F1-score levels of 97.83%, 98.12%, 97.46%, 98.72%, and 97.65%, respectively, outperforming the existing alternative models, including CNN, FSL, and ATCNN. It can reduce the volume of input data and accelerate the learning process by approximately 30%, resulting in improved efficiency and performance. The model's performance in classifying normal and abnormal ECGs is satisfactory. Novelty: The uniqueness of this work lies in the integration of three feature selection methods —GA, SVM, and PCA —into a CNN-LSTM network for ECG analysis. Unlike existing models, the proposed method can strike a good balance between classification ability and complexity. It has great potential for real-time ECG monitoring, AI-aided medical check-ups, and clinical decision support systems for cardiovascular medical care.