| 摘要: |
| 目的 探讨单纯自回归滑动平均（autoregressive integrated moving average，ARIMA）模型与ARIMA和非线性自回归（nonlinear autoregressive，NAR）组合模型在细菌性痢疾预测中的应用。方法 利用江苏省2004年1月至2015年2月的细菌性痢疾数据作为拟合样本，以2015年3月至2016年5月的数据作为预测样本；建立的模型分别为单纯ARIMA模型和ARIMA-NAR组合模型，然后根据2个模型的平均绝对误差（mean absolute error，MAE）、均方误差（mean square error，MSE）和平均绝对百分比误差（mean absolute percentage error，MAPE）比较模型的效果，其值越小模型效果越好。结果 在模型的拟合阶段，单纯ARIMA模型的MAE、MSE和MAPE分别为0.177 5、0.081 4和0.184 7，ARIMA-NAR组合模型分别为0.094 1、0.029 5和0.104 6。在模型的预测阶段，单纯ARIMA模型的MAE、MSE和MAPE也分别大于ARIMA-NAR组合模型。结论 ARIMA-NAR组合模型对于江苏省细菌性痢疾发病率时间序列的预测效果优于单纯ARIMA模型。建议尝试使用ARIMA-NAR组合模型预测细菌性痢疾的发病率。 |
| 关键词: 自回归滑动平均模型 非线性自回归模型 神经网络 时间序列 细菌性痢疾 预测 |
| DOI：10.16781/j.0258-879x.2017.10.1315 |
| 投稿时间：2017-03-09修订日期：2017-06-22 |
| 基金项目:江南大学自主科研青年基金（JUSRP11569），江南大学公共卫生研究中心项目（JUPH201508）. |
|
| Application of ARIMA-NAR combined model in predicting bacillary dysentery |
| WANG Ke-wei,LI Jin-ping,DENG Chao,WU Yu,WU Min-chen* |
(Department of Epidemiologic and Health Statistics, Wuxi Medical College, Jiangnan University, Wuxi 214122, Jiangsu, China
*Corresponding author) |
| Abstract: |
| Objective To explore the application of autoregressive integrated moving average (ARIMA) model,and ARIMA combined nonlinear autoregressive (ARIMA-NAR) model in predicting bacterial dysentery (BD) incidence. Methods Data of BD monthly incidences from Jan. 2004 to Feb. 2015 in Jiangsu Province were used as fitting samples, the 15-month data from Mar. 2015 to May 2016 were used in the prediction phase. ARIMA model and ARIMA-NAR model were established and the effects of two models were compared according to mean absolute error (MAE), mean square error (MSE), and mean absolute percentage error (MAPE), in which lower values suggested higher prediction accuracy. Results In the fitting phase, the MAE, MSE and MAPE of the ARIMA model were 0.177 5, 0.081 4 and 0.184 7, respectively, while those of the ARIMA-NAR model were 0.094 1, 0.029 5 and 0.104 6, respectively. In the prediction phase, the MAE, MSE and MAPE of the ARIMA model were significantly higher than those of the ARIMA-NAR model. Conclusion ARIMA-NAR combined model is superior to ARIMA model in predicting the time series of BD incidence in Jiangsu Province, suggesting that ARIMA-NAR model can be used to predict the incidence of BD. |
| Key words: autoregressive integrated moving average model nonlinear autoregressive model neural networks time series bacterial dysentery prediction |
