Redes Bayesianas para auxiliar no diagnóstico da COVID-19

• DOI: https://doi.org/10.22533/at.ed.9282428045

• Palavras-chave: Covid-19; K2; Naive Bayes; NB; Redes Bayesianas; SARS-Cov2; TAN; Tree Augmented Naive Bayes.

• Keywords: Covid-19; K2; Naive Bayes; NB; Bayesian Networks; SARS-Cov2; TAN; Tree Augmented Naive Bayes

• Abstract: The COVID-19 outbreak caused by the SARS-Cov2 coronavirus was a serious and urgent global concern. Using Bayesian Networks to aid in diagnosis would be beneficial for making quick decisions about treatment and isolation needs, as well as allowing to determine which characteristics presented by suspected cases of infection are the best predictors of a positive diagnosis. In this study, the objective was to apply different learning algorithms for Bayesian Networks, namely: K2, Tree augmented Naive Bayes (TAN) and Naive Bayes (NB). The algorithms were trained with a dataset available on the Kaggle website with information about Covid-19. The results obtained by the Bayesian classifiers (Categorical NB, TAN and K2) were able to efficiently deal with the COVID-19 target attribute classification task, presenting a similar performance in terms of accuracy, precision and recall reaching values greater than 97 \% on all measures. Furthermore demonstrated a high ability to identify the most relevant variables for a positive diagnosis. Finally, based on the studies, the effectiveness of Bayesian networks to assist in the diagnosis of COVID-19 was evidenced.