Predicting COVID-19 disease outcome and post-recovery conditions using machine learning

Abstract

With COVID-19 still running rampant across the world, accurate diagnosis of pa tients and proper management of medical resources is paramount in order to deliver proper care to those that need it most. In order to do this, prediction models with the help of various machine learning algorithms are being developed across the world. Each may deal with certain variables that help predict the disease outcome, such as comorbidities, symptoms, age, sex, etc. Some models have also been made to help predict the chances of a COVID-19 patient in developing lasting medical conditions post recovery. The goal of this research then, is to create a model that takes all the aforementioned dimensions into account and create a prediction model with the three timelines in mind. It is a model that will predict if a person has contacted COVID-19 based on the preliminary symptoms they show (Timeline 1), predict the chances of a COVID-19 patient developing more serious symptoms based on their medical history (Timeline 2) and also predict the chances of a patient developing post-recovery conditions arising after recovering from COVID-19 (Timeline 3). To accomplish this, we use three machine learning algorithms – Random Forest, Na¨ıve Bayes and K-nearest Neighbors. For implementation and testing of the model, data on COVID-19 patients is split into train and test sets and fit over the aforemen tioned algorithms. Their performance are then evaluated. Specific features of the dataset also analyzed at a deeper level in order to gain a better understanding of how the virus behaves in certain conditions. Having such a model in place will not only help us direct medical resources to patients that need the most attention, but will also provide a clearer understanding of the nature of the virus and how it affects a specific patient.