Forecasting COVID-19 Mortality Rates: A Comparative Study of utoregressive Integrated Moving Average and Neural Network Models

Accurate forecasting of infectious disease incidence is essential for timely intervention and effective government planning. This paper presents a comprehensive study comparing various forecasting models for daily COVID-19 mortality rates in Italy. The models evaluated include the autoregressive integrated moving average (ARIMA) model and three neural network-based models: backpropagation neural networks (BPNNs), radial basis function neural networks (RBFNNs), and Elman recurrent neural networks (ERNNs). RBFNN demonstrated superior performance with the lowest mean absolute error (MAE), mean absolute percentage error (MAPE), and mean square error (MSE), outperforming ARIMA and other neural networks by better capturing non-linear patterns in mortality data. The models’ performance ranking from best to worst was RBFNN, ERNN, BPNN, and ARIMA. These results underscore the effectiveness of neural network models, particularly RBFNN, in accurately forecasting COVID-19 mortality rates. The implications of these findings are significant for public health policy. The improved accuracy of RBFNN in short-term mortality prediction provides valuable insights for pandemic response planning, enabling health authorities to make informed decisions on resource allocation, public health advisories, and emergency preparedness. This study contributes to the literature on infectious disease modeling by demonstrating the advantages of neural networks over traditional statistical methods and offering practical guidance for selecting forecasting models in epidemic scenarios. Our evaluation of forecasting methods thus provides a critical foundation for enhancing predictive accuracy in disease incidence and supporting more responsive public health management.