Cerebral stroke is a critical neurological disorder where early and proper risk forecasting is necessary to minimize death and permanent disability. Conventional diagnostic methods can be manual and time consuming which restricts their applicability in the timely intervention. In the present paper, a machine learning framework based on stacking ensembles is suggested to use in the prediction of cerebral strokes. The framework combines the use of CatBoost, Decision Tree, and Random Forest as base learners to utilize their relative predictive advantages. An end-to-end machine learning pipeline is built that involves the pre-processing of data, features encoding, and normalization, minimization of class imbalance with the help of the Synthetic Minority Over-sampling Technique (SMOTE), applied to the training data only to avoid data leakage. Eleven machine learning models were first tested on an imbalanced Kaggle stroke dataset and the best models were chosen in building an ensemble, in a performance comparative analysis. As it is experimentally proved, the proposed stacking ensemble is better than the individual classifiers and results in the overall prediction accuracy of 98.64 percent and the enhanced precision, recall, and F1-score. In order to achieve practical usefulness, the trained model is implemented in the form of a web-based application, which can make real-time predictions of the risk of stroke based on patient demographic and clinical characteristics. The suggested framework has a promising future as a dependable decision-support platform of smart healthcare systems and prompt stroke risk detection.