Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, underscoring the importance of accurate prediction models for early intervention. This paper proposes a stacking ensemble model that integrates XGBoost, LightGBM, and Random Forest in the base layer, while employing Logistic Regression as the meta-level classifier to enhance heart disease risk prediction. The experimental results demonstrate that among traditional machine learning models, Decision Tree (DT) achieved the highest accuracy (86.57%), while Logistic Regression (LR) had the lowest (79.79%). However, after applying the stacking ensemble, LR's accuracy improved significantly to 91.00%, outperforming all other models, while DT only improved slightly to 87.10%. The findings highlight that stacking effectively addresses LR’s limitations in handling nonlinear data, significantly enhancing Precision, Recall, and AUC metrics. These results suggest that ensemble learning can provide a more robust and accurate framework for early heart disease risk assessment, contributing to improved clinical decision-making.