Heart disease is a leading global cause of mortality, emphasizing the urgent need for early and accurate diagnostic methods. However, traditional diagnostic tools often lack the scalability and precision to meet diverse healthcare demands. We propose a robust ensemble-based machine learning framework for heart disease prediction that combines advanced data preprocessing, feature selection, and model optimization techniques to address this. This study utilized a consolidated dataset of 1,888 records from five heart health databases. First, systematic preprocessing addressed missing values and standardized features. Next, a hybrid feature selection strategy, integrating Univariate Selection, Correlation Coefficient Analysis, and Mutual Information, reduced the feature set from 14 to 11 key indicators. Subsequently, we evaluated 12 machine learning models on all and reduced feature sets. Finally, our ensemble model, combining Random Forest, Decision Trees, XGBoost, CatBoost, and Extra Trees, achieved an outstanding accuracy of 98.12±1.20. Thus, our results demonstrate that integrating multiple classifiers with optimized feature selection significantly enhances predictive accuracy and model reliability. This scalable framework can transform early heart disease detection, improving outcomes across diverse healthcare settings