Bioaerossol alergênico: redes neurais a serviço da saúde pública

Abstract

Pollen is one of the main allergenic bioaerosols and has been gaining prominence over the years due to the growing population with allergic or related problems. Given its relevance to public health, measuring suspended pollen levels is a tool that can assist both affected individuals and healthcare professionals during periods of high pollination. As it is a reproductive structure involved in a complex biological cycle, linear models cannot accurately explain pollen behavior. Therefore, nonlinear models, such as artificial neural networks, are necessary to address this issue. This paper proposes the application of an artificial neural network to forecast atmospheric pollen using data from Southeastern Europe (provided by the authors of the Serbian-Croatian RealForAll project). To this end, using the Python programming language and the PyTorch library, Multi-Layer Perceptron (MLP) neural networks with varying numbers of hidden neurons and hidden layers were tested. Various statistical tests were applied to select the independent variables. The models with the lowest error were selected to predict the test set, with mean absolute error and root mean square error as the evaluation criteria. The objectives were achieved, but the MLP models performed worse when compared to other models applied to the same dataset (PollenNet, SM, and RM). Nonetheless, they show potential for reliable predictions of atmospheric pollen and its concentration peaks in their respective applications: public health and agriculture.