Machine Learning Models with Hybrid Ensemble Strategies Using Feature Engineering of Heart Disease Prediction
Abstract
Cardiovascular disease (CVD) is the number one cause of death worldwide, leading to millions of deaths each year and placing heavy pressure on health systems. Early diagnosis and management of heart disease is critical to optimize patient care and prevent adverse results. Many machine learning algorithms have been used in the prediction of heart disease, but these models tend to exhibit suboptimal performance in clinical practice because of feature redundancy, class imbalance, overfitting, and poor generalization. In this paper we proposed a hybrid stacked ensemble framework integrating twelve heterogeneous machine learning classifiers: KNN, Random Forest, Extra Trees, Gradient Boosting, LightGBM, AdaBoost, Balanced Random Forest, Naive Bayes, Logistic Regression, and Quadratic Discriminant Analysis, with a Logistic Regression meta learner as an alternative to overcome these issues. We will use Recursive Feature Elimination to identify the most informative clinical features and also employ cost-sensitive learning and Balanced Random Forest to mitigate class imbalance. To prevent unfair evaluation, the dataset is split into training, validation, and testing subsets. Hyper parameters are optimized via randomized search with 10-fold cross-validation. The experimental data have shown that the proposed framework has an F1-score of 0.96 and an accuracy of 0.97 and is a better performer than individual classifiers. And this is validated by confidence interval analysis, ablation studies, and statistical significance testing that prove the robustness and effectiveness of the proposed approach. The proposed framework is clinically feasible, scalable, and robust for heart disease prediction and allows us to implement it in healthcare analytics to perform real-time health information retrieval and in intelligent clinical decision-support systems.