Optimizing Ensemble Learning for Accurate Identification and Prognostic Evaluation of Cardiovascular Disease

Cardiovascular disease, a leading cause of global mortality, requires effective identification of key risk factors for accurate diagnosis. This study introduces a novel dynamic ensemble method, GMM-based Dynamic Ensemble Learning Optimization, utilizing a Multi-Objective Evolutionary Algorithm based on Decomposition, specifically designed for imbalanced datasets. By integrating an evolutionary resampling technique inspired by the Gaussian mixture model (GMM), the proposed approach addresses diversity, classifier performance, and the number of classifiers to generate optimal datasets. The method is evaluated on six heart datasets, exhibiting varying imbalance rates, and compared against established ensemble learning algorithms, consistently demonstrating superior performance across multiple evaluation metrics. Statistical significance tests, including McNemar's and Wilcoxon tests, further validate the findings. Furthermore, the application of Survival Analysis using Kaplan-Meier Estimates on a heart failure dataset enhances our understanding of heart disease prognosis. In conclusion, this research contributes to accurate heart disease prediction and prognosis assessment by leveraging innovative data mining and ensemble learning techniques.

Keywords:

Ensemble learning , Evolutionary multi-objective optimization , Bootstrap , Diversity , Heart disease