COVID-19 RISK CALCULATOR

Abstract

COVID-19 has been studied broadly and it has shown a very variant response depending on each individual, there are some factors that can make the disease more or less complex on someone infected. Machine learning models as decision trees have been used for short term prediction of mortality for different conditions such as chronic obstructive pulmonary disease. Vulnerability indexes are important during a pandemic to take especial care of the people who are more susceptible to worst outcomes such as death in case of infection. They are also important so the healthcare system can estimate the installed capacity beforehand and prepare for the upcoming infected patients. This project was designed for use by people in Latin America, especially in Colombia, the data used in this study was taken from Mexico’s Open Data from the General Directorate of Epidemiology. This article presents a machine learning model based on decision trees ensembles to predict the probability of someone dying because of the infection, using variables such as comorbidities, sex, gender and other individual conditions. The type of decision tree used was a LightGBM with Bayesian parameter optimization, subsequently calibrated using a sigmoidal function. The importance of each of the variables for the model was evaluated.The performance of the model was evaluated using the AUC ROC, and a result of 0.89 was obtained. This model was later used to develop a mortality calculator to assist health care workers and individuals in making decisions that can help them during the COVID-19 pandemic.