Boosting Ensemble Machine Learning Approach for Covid-19 Death Prediction

Abstract

It is critical for physicians to correctly classify patients during a plague and determine who deserves minimal health assistance. Machine learning methods have been presented to reliably forecast the severity of COVID-19 disease. Previous research has often tested different machine learning algorithms and evaluated performance under different methods. It may be necessary to try several combinations of machine learning algorithms to discover the optimal prediction model to get the best results. This research aimed to train boosting ensemble algorithms and Artificial Neural Networks (ANN) and choose the model that best predicted how long patients would survive a Covid19 infection. The dataset for this study was obtained through kaggle.com. It contains blood samples from 4313 patients and is retrospectively evaluated to find relevant measures of overall mortality. Out of 48 parameters, only 16 selected parameters were considered using the information gain weight for each parameter. 5-fold cross-validation was employed on the training data set, and Receiver Operating Characteristic (ROC) curves were created to verify better the prediction algorithms' performance independent of the algorithm choice criteria. The models XGBoost, CatBoost, and LightBGM achieved an accuracy of 98%, AdaBoost 96%, and 93% for ANN, respectively, implying that ANN has lower accuracy than boosting approaches.