The COVID-19 pandemic posed major challenges for medicine not only in patient care, but also in preparing health responses and managing public and private resources. Artificial intelligence (AI) is a new technology capable of analyzing complex databases to generate, among other outputs, predictions of future data.

We describe the development and prospective application of a machine learning model used to predict the number of SARS-CoV-2 infections using data from January 2021 to January 2022.

Development and use of an Artificial Neural Network (ANN) applied during the pandemic in a municipality in Serra Gaúcha-RS, after data cleaning, balancing, and segmentation into 74 neighborhoods/administrative regions according to the municipality’s geographic division plan. Variables included clinical, demographic, and socio-spatial data, including SINAN data, provided by hospitals and the city government. Models used: Multiple Linear Regression, Decision Tree Regressor, Random Forest Regressor, and a Multilayer Perceptron (MLP) ANN. The study was approved by the Research Ethics Committee (Opinion 4.587.313).

A total of 74,615 test results were used (24,147 in 2020 and 50,468 in 2021), with demographic, clinical, and socio-spatial variables. The MLP neural network had the best predictive performance among the tested models. The metric used to evaluate forecast accuracy was Mean Absolute Error (MAE). In August, the study showed 100% neighborhood coverage, with maximum errors of approximately 2.1 cases and values below 0.5 in central areas, demonstrating high predictive accuracy using 80% of the data for training and 20% for testing. Predictions were then made once per week, with the best results when forecasting incidence 30 days ahead.

The forecasts led to an estimated savings of R$118,996.80 for the state treasury in purchasing diagnostic tests. In addition, they helped managers adapt hospital sectors to create new Intensive Care Units (ICUs). A hospital with 2 ICUs prepared up to 8 ICUs during the pandemic.

The use of AI demonstrated not only predictive accuracy for COVID-19 cases during the pandemic, but also supported hospital management, rational resource use, and important public health measures for an effective response to the COVID-19 pandemic emergency.