Abstract
The medical community demands accurate predictive models for early heart disease diagnosis because heart disease remains a significant worldwide health concern. Deep learning research presents a predictive system for heart disease that uses K-mode clustering to optimize data preparation and implements Harris Hawks Optimization (HHO) for essential feature selection. The Cardiovascular Disease dataset of 70,000 patient records with many clinical parameters was used to develop model training and validation. The accuracy of cardiovascular disease prediction depends on neural networks and other deep learning architectures which analyze patient risk factors to determine disease development. The model operates efficiently using precision, recall, accuracy and the AUC score to evaluate its performance while utilizing the ROC curve. The developed feature selection method applying HHO increases model efficiency while maintaining prediction capabilities by eliminating unneeded features. The Gated Recurrent Unit (GRU) model achieved the highest accuracy of 88.03% among all the tested frameworks. Deep Learning methodologies integrated with advanced feature selection demonstrate high effectiveness in early detection of cardiovascular disease. It leads to efficient diagnostic solutions for health applications that scale across various systems.