<![CDATA[The increasing availability of Electronic Medical Records (EMR) presents significant opportunities for improving clinical decision-making through data-driven approaches. Nevertheless, the intricacy, heterogeneity, and high degree of dimensionality of EMR data pose challenges for conventional statistical techniques and single machine learning (ML) models in accurately predicting patient conditions, particularly disease severity. This study aims to develop and evaluate a stacking-based framework for ensemble learning for disease severity classification using EMR data. The proposed approach integrates multiple ML models, including LR, SVM, K-NN, RF, and NB, using a meta-learner to enhance predictive performance. The dataset comprises 4,412 patient records with clinical and laboratory features, and the assessment is carried out using accuracy, precision, recall, and F1-score metrics. The findings of the experiment demonstrate that the stacking-based ensemble model outperforms all individual models, achieving an accuracy of 0.790, a recall of 0.660, a precision of 0.770, and an F1-score of 0.710. The model also demonstrates improved capability in identifying severe cases while maintaining a balanced performance across evaluation metrics. These results verify that ensemble learning, particularly stacking, provides a more robust and generalized solution for handling complex EMR data. This research helps the development of ML in healthcare by demonstrating the efficiency of stacking-based ensemble frameworks in improving disease severity classification, thereby supporting more reliable and accurate clinical decision-making systems.]]>
