Abstract
BACKGROUND: The research shows the impact of land use forms and population density on the spatial variation of COVID-19 cases and deaths. Selected elements of land use that influence the number of COVID-19 cases and deaths were explored using an analysis of public data sets from 380 administrative units in Poland, which covered the entire pandemic period (from March 2020 to June 2023). This association has yet to be systematically elucidated with a machine learning method. METHODS: A machine learning model based on the structure of a multilayer artificial neural network was applied. Three independent models were developed to forecast COVID-19 cases and deaths and the case fatality ratio based on the type of county, population density and selected forms of land use (agricultural land, forest land, built-up urbanized, recreational, residential, transport and industrial). A global sensitivity analysis calculations were performed. RESULTS: The results show clear differences in COVID-19 rates between urban and land counties, which well reflects the impact of land use. The best results for COVID-19 case forecasting were obtained with the model consisting of 12 neurons in the hidden layers and the hyperbolic tangent activation function for the test set: R(2) = 0.977, MAE = 1596.8 people, and RMSE = 4158 people. The highest consistency of the COVID-19 death forecasts was obtained with the model consisting of 11 neurons and the sigmoid activation function in the hidden layers - for the test set: R(2) = 0.984, MAE = 26.9 people, and RMSE = 42.3 people. CONCLUSION: The results demonstrate a high-quality model for predicting viral infections, utilizing the three groups as input data (environmental, socioeconomic, and built environment) and employing artificial intelligence methods. Therefore, it may be a useful tool for analyzing the spread of epidemics. This approach can help national authorities and local governments to mitigate the high risk of infection and mortality in the future. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12889-026-27005-z.