COVID-19 predictions with Gradient Boosting Machine

Abstract

In this paper I have attempted to compare an efficiency and effectiveness of an Ensemble learning techniques, based on binary classification of SARS-CoV-2 disease. Boosting method and its applications will be the core of this thesis. I will compare various Gradient Boosting derivatives; represent their features, benefits and drawbacks. Also I will compare them with Neural Network. I have carry out my results and trained each model constructed on more than 3 millions COVID-19 test cases, over 300k of which had committed positive result. My results had illustrated the convenience and weight of Boosting methods within the framework of binary classification on categorical data, which means that attributes (variables) can take one of the limited number of possible values. Due to simplicity and productivity these methods these methods can be used in practice to recognize potential COVID cases and the symptoms that are influencing the decision of the model.