Enhancing Heart Disease Prediction through Multi-Modal Data Integration: A Hybrid CNN-GA Approach

Abstract

—Heart disease is among the most prevalent causes of mortality, and arrhythmia contributes a large segment; therefore, cost-effective, accurate, non-invasive diagnostic methodologies should be used. Structured clinical data-driven prediction models cannot fully capture the complex temporal–frequency features of Electrocardiogram (ECG) signals, while overfitting due to class imbalance and poor hyperparameter tuning has generally hindered effective implementation of machine learning models. To remedy these limitations, we propose a hybrid framework combining Convolutional Neural Networks (CNN) and a Genetic Algorithm (GA) for ECG-based heartbeat classification. The ECG signals are transformed into spectrogram images using Short-Time Fourier Transform (STFT), enabling effective feature extraction. The GA is used to optimize key CNN hyperparameters to improve classification performance. It uses stratified data split (70% for training, 15% for validation and 15% for testing) and class balancing techniques. The dataset used is the MIT-BIH Arrhythmia Database, and the model achieves state-of-the-art performance (98.21%), outperforming the baseline CNN (92%), pre-trained models such as ResNet50 and InceptionV3 (p = 0.03). Strong generalization is shown with five-fold cross-validation (98.1% ± 0.45). Combining evolutionary optimization with multi-modal data contributes to a strong, clinically relevant method for classifying arrhythmias.