Integrating Nursing Perspectives into Cardiovascular Disease Prediction Using Hybrid Classification Models and Metaheuristic Optimization with DOA, ARO, and PO Algorithms

Abstract

This study proposes a cardiovascular disease (CVD) prediction model based on the XGBoost classifier, enhanced through metaheuristic-based hyperparameter optimization. A real-world dataset comprising 70,000 patient records including attributes such as age, blood pressure, cholesterol levels, and lifestyle factors was used for model development. Three recent metaheuristic algorithms Dingo Optimization Algorithm (DOA), Artificial Rabbits Optimization (ARO), and Political Optimizer (PO) were employed to optimize critical XGBoost hyperparameters such as learning rate, maximum depth, and number of estimators. The dataset was divided using a stratified 80:20 train-test split. Performance was evaluated using accuracy, precision, recall, and F1-score metrics. Among all configurations, the XGBoost model optimized via ARO (XGAR) achieved the best performance, with a training accuracy of 0.995, a testing accuracy of 0.678, and an overall average accuracy of 0.900. These findings highlight the effectiveness of combining XGBoost with metaheuristic optimization for early CVD risk prediction and potential integration into preventive healthcare frameworks.