Machine Learning: A Tool for Formulating Tracking Strategies for Covid-19 Pandemic in Colombia

Abstract

The Covid-19 pandemic has greatly affected the health conditions of the population as well as the economies around the world. In some countries, health systems have been overwhelmed, since in most cases they were not prepared to assist the large number of people who were affected . In the absence of vaccines and treatments of proven efficacy, the most obvious measure is the lockdown. An initial lockdown of five months was one of the first measures taken by the Colombian government to impede the spread of the virus. However, this measure did not prevent the country from becoming the fifth in number of infected people worldwide. With a population of about 8 million people and the social dynamics and mobility typical of a capital city, Bogotá has high rates in the statistics of incidence, effective reproduction and mortality. It is assumed that after reaching herd immunity, the incidence of the virus diminishes. But the current priority in the country is to focus on the scarce resources on the sites and the communities most prone to contagion. This document presents the use of machine learning-based models to identify these potential nuclei. A clustering model was constructed using the Simple K-means algorithm to identify clusters of people who are most susceptible to contracting the virus. Four algorithms were used for determining the value of the effective reproduction number. The best performing algorithm was linear regression with a correlation coefficient of 0.753 and artificial neural networks with 0.774.