<p>Cardiovascular disease (CVD) is a major global health concern, demanding accurate predictive models to aid preventive healthcare strategies. Heart failure, stroke, and coronary artery disease are among the disorders that fall under the umbrella term of cardiovascular disease (CVD). Leveraging the Harris Hawks Optimization (HHO) algorithm in conjunction with deep belief networks (DBNs) aims to improve CVD risk prediction accuracy. Harris Hawks Optimization (HHO) draws inspiration from the cooperative behavior of Harris’s hawks in nature, providing an efficient metaheuristic search algorithm for optimization problems. Integrating HHO into machine learning frameworks enhances the exploration and exploitation of search spaces, leading to improved model performance and convergence rates, particularly in deep learning tasks like feature selection and hyperparameter tuning. Powerful generative models called deep belief networks (DBNs) are made up of several layers of latent, stochastic variables. Restricted Boltzmann machines (RBMs) serve as building blocks in the training process of deep belief networks (DBNs), facilitating the unsupervised pre-training of hidden layers. Leveraging RBMs within DBNs enables the extraction of hierarchical representations, enhancing the network’s ability to learn intricate patterns and improve predictive performance in complex data settings. They leverage unsupervised learning techniques to extract intricate patterns and hierarchical representations from complex data, making them ideal for tasks such as feature learning and classification in machine learning research. This study introduces innovative algorithms, including the correlation-based weighted compound feature generation (CWCFG) technique, to enhance the optimization process of HHO. Comparative analysis against traditional machine learning models and rule-based firefly optimizer (RBFO) and Grey Wolf Optimizer (GWO) with the state-of-the-art deep learning techniques demonstrates the efficacy of the CWCFG-HHO-DBN model. Additionally, an in-depth feature importance analysis identifies key predictors, enriching the model’s interpretability. The research conducts a comprehensive evaluation of the proposed model, employing various performance metrics such as accuracy, precision, recall, and F-measure. With a remarkable accuracy of 97.19%, the HHO-DBN model shows promise in enhancing CVD risk prediction. The findings underscore its potential in personalized medicine, facilitating tailored interventions for high-risk individuals. Future directions include refining the algorithm and expanding its application in healthcare settings.</p>