Utilização de redes neurais artificiais para prever casos de raiva bovina nos estados brasileiros.

Abstract

Bovine rabies is an endemic zoonosis characterized by high lethality in Brazil, leading to significant economic losses in the livestock sector due to the mortality of production animals. The disease is primarily transmitted by the hematophagous bat Desmodus rotundus. Despite prophylactic actions coordinated by the Programa Nacional de Controle da Raiva dos Herbívoros (PNCRH) of the Ministério da Agricultura e Pecuária (MAPA), the predictability of its occurrence remains limited. This study aimed to evaluate the predictive capacity of Artificial Neural Networks (ANNs) in estimating the annual occurrence of bovine rabies nationwide and identifying risk areas. The methodology utilized a Multilayer Perceptron (MLP) feedforward ANN model, trained on historical time- series data of cases reported by MAPA. Performance was assessed using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination (R2). The results demonstrated satisfactory model performance, with most states achieving an R2 above 0.98, confirming the efficacy of the approach in capturing the non-linear dynamics of the epidemiological series. Projections through 2025 suggest a trend of stability or slight reduction; however, the highest concentration of predicted cases is expected in the South and Central-West regions, indicating continued risk. ANNs prove to be a promising and reliable tool for epidemiological surveillance, enabling the strategic direction of preventive measures.