Detection and risk assessment of COVID-19 through machine learning

Abstract

COVID-19, also known as coronavirus disease, is caused by the SARS-CoV-2 virus. People infected with COVID-19 may show a range of symptoms from mild to severe, including fever, cough, difficulty breathing, tiredness, and nasal congestion, among others. The goal of this study is to use machine learning to identify if a person has COVID-19 based on their symptoms and to predict how severe their illness might become. This could lead to outcomes like needing a ventilator or being admitted to an Intensive Care Unit. The methods used in this research include Artificial Neural Networks (specifically, Multi-Layer Perceptrons), Classification and Regression Trees, and Random Forests. Data from the National Epidemiological Surveillance System of Mexico City was analyzed. The findings indicate that the Multi-Layer Perceptron model was the most accurate, with an 87.68% success rate. It was best at correctly identifying COVID-19 cases. Random Forests were more effective at predicting severe cases and those requiring Intensive Care Unit admission, while Classification and Regression Trees were more accurate in identifying patients who needed to be put on a ventilator.