Comparação de algoritmos de aprendizado de máquina para predizer futuras cepas do vírus da influenza

Abstract

Every year, efforts are made to produce vaccines that are effective against the influenza virus. However, in order for a vaccine to have the desired effect, in-depth studies on pos- sible virus mutations are necessary. Although there are several methods in the literature for predicting new strains of the influenza virus, only a few utilize artificial intelligence as a tool in this process. In light of this, this study proposes the implementation of learning algorithms to compare their performance in predicting new strains and to compare them with existing models in the literature. The learning models were implemented in Python, using external libraries such as Random Forests, Extra Trees, Naive Bayes, and Decision Trees. After implementation, data from H1N1 and H3N2 strains were collected from a database and organized in FASTA files in chronological order. The execution of all algo- rithms revealed that the Random Forests model had a higher accuracy than the others. When compared to the models in the literature, it was found that the Random Forests algorithm performed well, second only to a model that utilizes predictive fitness.