The first case of the novel coronavirus in Brazil was notified on February 26, 2020. After 21 days, the first case was reported in the second largest State of the Brazilian Amazon. The State of Pará presented difficulties in combating the pandemic, ranging from underreporting and a low number of tests to a large territorial distance between cities with installed hospital capacity. Due to these factors, mathematical data-driven short-term forecasting models can be a promising initiative to assist government officials in more agile and reliable actions. This study presents an approach based on artificial neural networks for the daily and cumulative forecasts of cases and deaths caused by COVID-19, and the forecast of demand for hospital beds. Six scenarios with different periods were used to identify the quality of the generated forecasting and the period in which they start to deteriorate. Results indicated that the computational model adapted capably to the training period and was able to make consistent short-term forecasts, especially for the cumulative variables and for demand hospital beds.

Resumo Objetivo: Relatar o produto de pesquisa e extensão universitária denominado Boletim COVID-PA, que apresentou projeções sobre o comportamento da pandemia no estado do Pará, Brasil. Métodos: Utilizou-se da técnica de inteligência artificial conhecida como ‘redes neurais artificiais’, para geração de 13 boletins com projeções de curto prazo baseadas nos dados históricos do sistema da Secretaria de Estado de Saúde Pública. Resultados: Após oito meses de projeções, a técnica gerou resultados confiáveis, com precisão média de 97% (147 dias observados) para casos confirmados, 96% (161 dias observados) para óbitos e 86% (72 dias observados) para ocupação de leitos de unidade de terapia intensiva. Conclusão: Esses boletins tornaram-se um instrumento útil para a tomada de decisão de gestores públicos, auxiliando na realocação de recursos hospitalares e otimização das estratégias de controle da COVID-19 nas diversas regiões do estado do Pará.

Modeling time series has been a particularly challenging aspect due to the need for constant adjustments in a rapidly changing environment, data uncertainty, dependencies between variables, volatile fluctuations, and the need to identify ideal hyperparameters. The present study presents a Framework capable of making projections from time series related to cases and deaths by COVID-19 in the Amazonian state of Pará, in Brazil. For the first time, deep learning models such as TCN, TRANSFORMER, TFT, N-BEATS, and N-HiTS were assessed for this purpose. The ARIMA statistical model was also used in post-processing for residual adjustment and short-term smoothing of the generated forecasts. The Framework generates probabilistic forecasts, with multivariate support, considering the following variables: daily cases per day of the first symptom, cases published daily, the occurrence of deaths, deaths published daily, and percentage of daily vaccination. The generated predictions are statistically evaluated by determining the best model for 7-day moving average projections using evaluating metrics such as MSE, RMSE, MAPE, sMAPE, r², Coefficient of Variation, and residual analysis. As a result, the generated projections showed an average error of 5.4% for Cases Publication, 8.0% for Cases Symptoms, 11.12% for Deaths Publication, and 4.6% for Deaths Occurrence, with the N-HiTS and N-BEATS models obtaining better results. In general terms, the use of deep learning models to predict cases and deaths from COVID-19 has proven to be a valuable practice for analyzing the spread of the virus, which allows health managers to better understand and respond to this kind of pandemic outbreak.

Neste estudo, foi explorado a aplicação de técnicas de Aprendizado Profundo (Deep Learning) para a classificação de plantações de coqueiro a partir de imagens coletadas por uma aeronave remotamente pilotada (RPA). Foi utilizado Rede Neural Convolucional para criação de um modelo que classifica imagens em duas categorias: coqueiro e não-coqueiro. O modelo obteve a acurácia de 88%, concluindo que técnicas de Deep Learning são ferramentas viáveis para a identificação de plantas em áreas de grande extensão, com potencial para melhorar a gestão de culturas agrícolas.