Feature Selection and Classification of Hypertension Using the Chaotic Golden Eagle Optimization Algorithm

Abstract

Hypertension is a major global public health problem and one of the leading risk factors for cardiovascular disease. Accurate classification and early identification are essential for effective treatment planning and long-term risk reduction. However, machine-learning models are often affected by redundant, irrelevant, or highly correlated features, which may reduce classification performance and increase computational burden. This study proposes a Chaotic Golden Eagle Optimization (CGEO) algorithm for feature selection and hypertension classification. The proposed method extends the original Golden Eagle Optimizer (GEO) by integrating a Chebyshev chaotic map to improve population diversity, strengthen the exploration–exploitation balance, and reduce premature convergence. The method was evaluated on a real hypertension dataset containing clinical and demographic attributes and was integrated with four classifiers: Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Decision Tree (DT), and Naïve Bayes (NB) under the same experimental settings. Model performance was assessed using accuracy, precision, recall, F1-score, and specificity over 30 independent runs. The experimental results showed that CGEO consistently outperformed GEO across all classifiers. Among the evaluated models, CGEO-SVM achieved the best overall performance, with an accuracy of 95.78 percent and a specificity of 96.20 percent, while reducing the number of selected features by approximately 50%. An additional controlled SVM-based benchmark against PSO, GWO, and GA further showed that CGEO achieved the most favorable overall trade-off in terms of accuracy, specificity, F1-score, feature-reduction rate, and runtime under the present experimental setting. Statistical testing further confirmed that the differences between GEO and CGEO were significant at the 0.05 level. In addition, although CGEO required slightly more computation time than GEO, the overhead remained acceptable relative to the gains in predictive performance, stability, and feature subset compactness. These findings indicate that incorporating chaotic dynamics into a metaheuristic framework can significantly improve feature selection and medical data classification performance.