The need for precise and simple diagnostic techniques is highlighted by the fact that cardiovascular diseases (CVDs) remain one of the major risks to world health. This study proposes a hybrid deep learning-based architecture for heart disease prediction using the publically accessible Heart Disease dataset, which includes 920 patient records and 13 significant clinical characteristics. The suggested model, known as Power Boost Ensemble, uses a stacking technique to merge four distinct base learners: Random Forest, Extra Trees, Gradient Boosting, and Logistic Regression. A Ridge Classifier serves as the meta learner in this configuration, gathering predictions from each base learner. With a test accuracy of 85% using 10-fold cross validation, the stacked ensemble exhibits good generalization and consistent performance across all significant assessment criteria. Shapley Additive Explanations (SHAP) are used to understand how the meta model develops its predictions in order to improve interpretability and clarity. The SHAP results show that the model’s conclusions are significantly influenced by important clinical parameters including ca (number of main vessels), cp (kind of chest pain), thal (thalassemia), and oldpeak (ST depression). All things considered, the Power Boost Ensemble offers a dependable, comprehensible, and reproducible approach to cardiac sickness prediction, making it appropriate for upcoming clinical applications based on artificial intelligence.

Deep Learning,Explainable AI,Multi- Layer Perceptron,Heart Disease,Federated Learning Models