Avaliação da metodologia de aprendizagem contínua na predição de casos de pacientes infectados com Covid-19

Abstract

COVID-19 is an infectious disease caused by the SARS-CoV-2 virus, which can lead to Severe Acute Respiratory Syndrome (SARS) and result in death. The use of Machine Learning (ML) can help predict and assist in decision-making in cases of infection that can become serious and lead to death. The traditional ML may have difficulties, due to the change in dynamics during the pandemic with the advancement of vaccination, resulting in loss of performance of the models. This happens because traditional ML is trained on historical datasets and is not updated during use. To deal with this type of problem, Continuous Learning (CA), one of the areas of ML, allows them to adapt to changes in the scenario, improving the performance and robustness of precisions. Variable engineering methods and different models were used, employing the CA methodology, with the aim of obtaining better performance and greater stability in the predictions of cases of cure or death based on the symptoms and comorbidity of the patients. Evaluating different models and methodologies with and without the use of continuous learning. The best model achieved an average AUC score of 0.740 with a standard deviation of 0.015. By comparing the experiments with and without the use of the AC methodology, it is possible to infer the potential impact on the performance of the models throughout their use.