Predicting COVID-19 Patient Outcomes: A Comparative Analysis of the Gradient Boosted Random Forest Classifier

Abstract

The COVID-19 pandemic has highlighted the limitations of traditional AI models in predicting patient health outcomes, particularly their inability to generalize across diverse populations and evolving viral strains. Existing Machine Learning (ML) and Deep Learning (DL) models often suffer from dataset bias, lack of adaptability, and performance degradation over time. This study proposes a self-adaptive ensemble learning framework that combines Gradient-Boosted Random Forest (GBRF), a Voting Classifier, and a Convolutional Neural Network (CNN) to enhance predictive robustness across heterogeneous COVID-19 patient datasets. Unlike conventional models, our approach dynamically adjusts feature importance based on real-time data, ensuring adaptability to new pandemic waves and emerging variants. Experimental results demonstrate that our GBRF model achieved 95.6% accuracy, outperforming static ML models and maintaining high precision and recall across geographically diverse patient cohorts. These findings underscore the potential of adaptive AI models in healthcare analytics, not only for COVID-19 but for future infectious disease outbreaks, providing a scalable solution for real-time clinical decision support and resource allocation.