Data Science and Artificial Intelligence in Predicting Severity of COVID-19 as a Risk Factor for Long COVID

Abstract

This work applies data science and machine learning techniques to predict the severity of COVID-19 (mild vs. severe cases) based on demographic information, comorbidities, and clinical symptoms, using public data from the Recife City Hall. A complete pipeline was developed, including preprocessing, feature engineering, and evaluation of classification models. The results show that the inclusion of clinical symptoms significantly improves the performance in severity classification. Logistic regression showed good performance and interpretability, allowing the identification of factors associated with severity, such as the number of comorbidities and respiratory symptoms related to oxygenation. As the dataset does not contain an explicit outcome of Long COVID, severity was discussed as a proxy for risk, according to evidence from the literature, highlighting that the relationship with persistent symptoms is not directly estimated in this database. Thus, the developed models can contribute to the early identification of patients with a higher probability of severe progression and, consequently, a higher risk of Long COVID.