Analysis of signs and symptoms of SARS-CoV-2 virus infection considering different waves using Machine Learning

In March 2020, the World Health Organization declared a world pandemic of COVID-19, which can manifest in humans as a consequence of virus infection of SARS-CoV-2. On this context, this work uses Data Mining and Machine Learning techniques for the infection diagnosis. A methodology was created to facilitate this task and can be applied in any outbreak or pandemic wave. Besides generating diagnosis models based only on signals and symptoms, the method can evaluate if there are differences in signals and symptoms between waves (or outbreaks) through explainable techniques of the machine learning models. Another aspect is identifying possible quality differences between exams, for example, Rapid Test (RT) and Reverse Transcription-Polymerase Chain Reaction (RT-PCR). The case study in this work is based on data from patients who sought care at Piquet Carneiro Polyclinic of the State University of Rio de Janeiro. In this work, the results obtained with the tests were used to diagnose symptomatic infection of the SARS-CoV-2 virus, based on related signals and symptoms, and the date of the initial of these signals and symptoms. Using the Random Forrest model, it was possible to achieve the result of up to 76% sensitivity, 86% specificity, and 79% accuracy in the results of tests in one contagion wave of the SARS-CoV-2 virus. Moreover, differences were found in signals and symptoms between contagion waves, in addition to the observation that exams RT-PCR and RT Antigen tests are more reliable than RT antibody test.