Severity classification and disposition prediction using ensemble learning for home-based patient management with adequate decision making

Home-based patient evaluation is crucial for determining whether patients require immediate transfer to emergency departments (ED) or can safely remain at home. Traditional methods relying on subjective clinical judgments can lead to inconsistencies, ED overcrowding, and increased healthcare costs. This study proposes a dual-phase support model integrating severity classification using the Emergency Severity Index and patient disposition prediction through a stacking ensemble learning (EL) approach. Our model combines Multilayer Perceptron (MLP), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) as base learners, with RF as the meta-learner. Additionally, we developed and compared a voting EL model using the same base learners. To further assess performance, we evaluated the individual models alongside deep learning approaches, including Long Short-Term Memory and Deep Convolutional Neural Networks. Model performance was assessed using a real-world dataset with accuracy and the area under the receiver operating characteristic curve (AUROC) as evaluation metrics. The proposed stacking ensemble model achieved 91.68% accuracy and an AUROC of 97.04%, outperforming individual models and the voting EL. Integrating severity classification with disposition prediction improved accuracy by 1.28%. The findings highlight the effectiveness of stacking ensemble learning in enhancing patient evaluation accuracy, optimizing healthcare resource allocation, and reducing unnecessary ED visits. Additionally, the combination of severity classification and disposition prediction contributed to improved model performance, demonstrating the advantage of integrating both tasks in a unified framework. Comparative analysis of different models provided valuable insights into performance tradeoffs, reinforcing the potential of ensemble learning for home-based patient assessment.