Enhanced Brain Stroke Prediction Using Boosting and Stacking Ensemble Learning

Abstract

Stroke is a major cerebrovascular disorder and one of the leading causes of mortality and long-term disability worldwide. Early identification of individuals at increased risk of stroke can support timely intervention and preventive decision-making. The increasing availability of structured clinical and demographic data has encouraged the development of machine-learning-based prediction models capable of identifying complex nonlinear relationships among patient characteristics. However, individual machine-learning algorithms may exhibit limitations when dealing with class imbalance, heterogeneous feature distributions, missing observations, and complex feature interactions. To address these challenges, this study proposes RXLM, a stacked ensemble framework integrating Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) for stroke prediction. The proposed framework incorporates missing-value handling, outlier treatment, categorical feature encoding, normalization, and training-set-specific oversampling, followed by independent hyperparameter optimization of the three base learners. Their predictive outputs are subsequently integrated through a stacked ensemble architecture to obtain the final prediction. The proposed framework is evaluated using accuracy, precision, recall, F1-score, area under the receiver operating characteristic curve (AUC), Cohen's kappa, and Matthews correlation coefficient (MCC). In the experimental setting, RXLM achieved an accuracy of 93.6%, precision of 82.7%, recall of 86.0%, F1-score of 84.3%, AUC of 96.5%, kappa of 0.799, and MCC of 0.800, outperforming the individual base learners across the majority of evaluation measures. These results demonstrate the potential benefit of combining heterogeneous tree-based learners through optimized stacking. Further validation using independent multicenter clinical datasets is required before clinical deployment.