Enhanced COVID-19 Classification Using Ensemble Meta-Algorithms on Chest X-ray Images

Ensemble learning is a commonly used approach in machine learning where multiple learners are trained and merged to solve the same problem. The main hypothesis is that the collective ability can offer improved predictive performance compared to that of individual learners. Two widely used techniques for ensemble learning are bagging and boosting, which help to reduce the variance of a single estimate by combining several estimates from different models. In this study, the effectiveness of bagging and boosting was investigated for the classification of COVID-19 patients based on their chest x-rays. For the bagging method, a model based on voting classifier was proposed, while the boosting methods used in the proposed models were AdaBoost and XGBoost algorithms. The proposed models demonstrated superior results in multiclass classification compared to individual machine learning methods. The proposed XGBoost-based model performs better than proposed AdaBoost and bagging-based models when it used for multiclass classification. The traditional bagging method achieved a classification recall score of 90.3%, while the proposed bagging model achieved a recall of 95.6%. When using the proposed AdaBoost and XGBoost-based models, we achieved recall scores of 94.0% and 96.2%, respectively. XGBoost demonstrated the best performance among all ensemble methods and individual base learners in all experiments. Across all experiments, XGBoost achieved the best performance compared to other ensemble methods and the individual base learners.