CLASSIFICAÇÃO DE DADOS DE COVID-19 USANDO REDE NEURAL CONVOLUCIONAL

Abstract

In the year 2020 the world was surprised by the emergence of a pandemic, identified as Covid-19. Starting in China and spreading across the globe in a short period of time with flu-like symptoms, its rapid transmission and evolution of cases surprised the medical community. During these two years, many doctors and scientists have struggled to understand the patterns of the disease, and one of the methods used was the use of neural networks that use artificial intelligence techniques to identify the patterns of the virus. To contribute to the research developed during the pandemic, this study aims to evaluate, via a case study, a 1D convolutional neural network for the classification of patients infected or not by the virus. Using a database made available by Hospital S ́ırio Libanês de S ̃ao Paulo. The specific objectives are to contextualize the Covid-19 pandemic scenario, introduce and conceptualize deep learning, describe and explain what a neural network is and the use of 1D convolutional neural networks, and finally apply the 1D convolutional neural network to the classification of Covid-19 data for outcome prediction. After training the neural network model developed, it was possible to see that the classification obtained an accuracy above 90\%, but when analyzing the minority class this value approaches 0, this is due to the large unbalance of the data. However, at the end of the development, it was found that the use of this database, which has a larger amount of negative data than positive, did not reach a sufficient hit rate for positive data, making it is evident that it is necessary to obtain more data of this class in order to improve the model's classification results