<p>Cardiovascular disease (CVD) is still a major cause of death globally, and hence the need for a precise and robust predictive model for early detection. While many machine learning models have been developed, most of these are based on single classifiers, simple validation techniques, or inadequate robustness analysis. To overcome these challenges, this paper presents an optimized stacking ensemble model for early cardiovascular disease prediction based on structured clinical data. The proposed model combines three diverse base models, namely Random Forest, XGBoost, and CatBoost, with a meta-model of Logistic Regression for enhanced predictive robustness and generalization. The proposed model was tested on the Heart Statlog Hungary Cleveland dataset, which has 1190 patient samples and 14 clinical features. A structured data preprocessing technique was designed, which included missing data treatment, outlier identification, and feature scaling to cope with the heterogeneity of the clinical data. Hyperparameters of individual base models were tuned using grid search. For testing the robustness of the model, the framework was tested on various train and test splits (70:30, 80:20, and 90:10) as well as cross-validation. The stacking model designed in this study performed better, with a maximum accuracy of 97.98%, precision of 98.57%, recall of 95.83%, F1-score of 97.18%, and AUC-ROC of 98.64%. The above results show that the optimized stacking ensemble model has better predictive reliability and fewer false negatives than the individual models. This study provides a reliable and practical decision-support system for the early diagnosis of cardiovascular disease.</p>