COVID-19 Mortality Prediction Using Machine Learning Techniques

Abstract

The COVID-19 pandemic sparked our research interest to explore and design a predictive model through Machine Learning algorithms to determine risk and mortality of COVID-19 admitted patients. Using a data set with over 90,000 patient admits and 20 clinical health features, this study aims to help prioritize care on patients that have a higher risk for COVID-19 based on their bill of health. The accuracy in predicting mortality rate was 96 percent on high performing models. Research methods included data mining using WEKA, Ensemble Learning Techniques with feature tuning on the the following algorithms: Navies Bayes, Decision Trees, K-Nearest Neighbor, Support Vector Machine (SVM), Random Forrest and Multilayer Perceptron (MLP). Tuning the models was achieved through feature selection, ranking, wrapping and filtering.