Wu Wei, An Shuyi, Guo Junqiao, Guan Peng, Ren Yangwu, Xia Lingzi, Zhou Baosen
Department of Epidemiology, School of Public Health, China Medical University, Shenyang 110122, China.
Liaoning Provincial Center for Disease Control and Prevention.
Zhonghua Liu Xing Bing Xue Za Zhi. 2015 Dec;36(12):1394-6.
To explore the prospect of nonlinear autoregressive neural network in fitting and predicting the incidence tendency of hemorrhagic fever with renal syndrome (HFRS) , in the mainland of China.
Monthly reported case series of HFRS in China from 2004 to 2013 were used to build both ARIMA and NAR neural network models, in order to predict the monthly incidence of HFRS in China in 2014. Fitness and prediction on the effects of these two models were compared.
For the Fitting dataset, MAE, RMSE and MAPE of the ARIMA model were 148.058, 272.077 and 12.678% respectively, while the MAE, RMSE and MAPE of NAR neural network appeared as 119.436, 186.671 and 11.778% respectively. For the Predicting dataset, MAE, RMSE and MAPE of the ARIMA model appeared as 189.088, 221.133 and 21.296%, while the MAE, RMSE and MAPE of the NAR neural network as 119.733, 151.329 and 11.431% respectively.
The NAR neural network showed better effects in fitting and predicting the incidence tendency of HFRS than using the traditional ARIMA model, in China. NAR neural network seemed to have strong application value in the prevention and control of HFRS.
探讨非线性自回归神经网络在中国内地肾综合征出血热（HFRS）发病趋势拟合及预测中的应用前景。
利用2004年至2013年中国HFRS的月报告病例序列建立ARIMA模型和NAR神经网络模型，以预测2014年中国HFRS的月发病率。比较这两种模型在拟合和预测效果方面的表现。
对于拟合数据集，ARIMA模型的平均绝对误差（MAE）、均方根误差（RMSE）和平均绝对百分比误差（MAPE）分别为148.058、272.077和12.678%，而NAR神经网络的MAE、RMSE和MAPE分别为119.436、186.671和11.778%。对于预测数据集，ARIMA模型的MAE、RMSE和MAPE分别为189.088、221.133和21.296%，而NAR神经网络的MAE、RMSE和MAPE分别为119.733、151.329和11.431%。
在中国，NAR神经网络在HFRS发病趋势的拟合和预测方面比传统的ARIMA模型效果更好。NAR神经网络在HFRS的防控中似乎具有较强的应用价值。
