de Araújo Morais Lucas Rabelo, da Silva Gomes Gecynalda Soares
Department of Statistics, Federal University of Bahia, Salvador, Bahia, Brazil.
Appl Soft Comput. 2022 Sep;126:109315. doi: 10.1016/j.asoc.2022.109315. Epub 2022 Jul 15.
The use of models to predict disease cases is common in epidemiology and related areas, in the context of Covid-19, both ARIMA and Neural Network models can be applied for purposes of optimized resource management, so the aim of this study is to capture the linear and non-linear structures of daily Covid-19 cases in the world by using a hybrid forecasting model. In summary, the proposed hybrid system methodology consists of two steps. In the first step, an ARIMA model is used to analyze the linear part of the problem. In the second step, a neural network model is developed to model the residuals of the ARIMA model, which would be the non-linear part of it. The neural network model was superior to the ARIMA when considering the capture of weekly seasonality and in two weeks, the combination of models with the capture of seasonality in two weeks provided a mixed model with good error metrics, that allows actions to be premeditated with greater certainty, such as increasing the number of nurses in a location, or the acceleration of vaccination campaigns to diminish a possible increase in the number of cases.
在流行病学及相关领域，使用模型预测疾病病例的情况很常见。在新冠疫情背景下，自回归积分滑动平均（ARIMA）模型和神经网络模型均可用于优化资源管理。因此，本研究旨在通过使用混合预测模型捕捉全球每日新冠病例的线性和非线性结构。总之，所提出的混合系统方法包括两个步骤。第一步，使用ARIMA模型分析问题的线性部分。第二步，开发神经网络模型对ARIMA模型的残差进行建模，这将是其非线性部分。在考虑捕捉每周季节性方面，神经网络模型优于ARIMA模型，并且在两周内，结合捕捉两周季节性的模型提供了一个具有良好误差指标的混合模型，这使得能够更确定地预先策划行动，例如增加某个地点的护士数量，或加快疫苗接种运动以减少病例数可能的增加。
