Classificador Naive Bayes Multinomial

Abstract

In this work, we present the theoretical basis of the Multinomial Naive Bayes machine learning classifier. Machine learning libraries such as Python’s sklearn were not used in the development of the algorithms of the Naive Bayes classifier, these algorithms were implemented using probabilistic arguments, the Bayes rule, and the hypothesis of independence between the variables. This allows Naive Bayes classifier implementation in any programming language. To test the classifier, we used data with fake and real news about COVID-19 in Brazil, compiled from posts on social networks, and the algorithms were implemented in Python and JavaScript programming languages. To verify the efficiency of the classifier, 100 simulations were performed to detect fake news, the results in both programming languages shows an accuracy level around 85%, precision of 91%, sensitivity of 80%, and F1-score of 85%.