α-Sutte、SARIMA及其组合模型在甲肝发病率预测中的应用效果比较

Comparison of α-Sutte, seasonal autoregressive integrated moving average model and their combination model for prediction of hepatitis A incidence

目的 比较α-sutte、SARIMA及其组合模型(SutteSARIMA)在甲肝发病率预测中的应用效果,为优化甲肝预测模型提供参考。方法 收集2004—2017年全国甲肝逐月发病率数据。以2004年1月—2017年6月数据作为训练集,2017年7—12月数据作为测试集。利用训练数据分别训练α-sutte、SARIMA及SutteSARIMA模型。利用训练模型预测2017年7—12月发病率,并与测试集比较。采用平均绝对误差百分比(Mean Absolute Percentage Error, MAPE),平均绝对误差(Mean Absolute Error, MAE),均方根误差(Root Mean Squared Error, RMSE)和平均误差率(Mean ErrorRate, MER)评价模型拟合及预测效果。结果 α-sutte、SutteSARIMA模型残差均在0附近波动。α-sutte、SARIMA及SutteSARIMA模型拟合的MAPE、MAE、RMSE、MER依次为7.68%、0.02、0.03、6.34%,12.10%、0.03、0.05、12.18%,7.72%、0.02、0.03、7.27%;预测的MAPE、MAE、RMSE、MER依次为17.13%、 0.02、0.02、16.17%,15.32%、0.02、0.02、15.54%,5.88%、0.01、0.01、5.80%。SARIMA(1,1,3)(2,1,1)12为最优SARIMA模型。结论 SutteSARIMA为最优模型,适于全国甲肝发病率的预测,值得推广至其他疾病的预测。

Objective To compare the effectiveness of α-Sutte Indicator, seasonal autoregressive integrated moving average(SARIMA) model and their combination model(SutteSARIMA) for prediction of hepatitis A incidence;to provide insights into optimization of predictive models for hepatitis A. Methods The monthly incidence of hepatitis A in China from 2004 to 2017 were collected, with data captured between Jan 2004 and Jun 2017 as training sets and data captured during the period between Jul and Dec 2017 as test sets. The α-Sutte, SARIMA and SutteSARIMA models were trained with training sets, employed for prediction of hepatitis A incidence during the period between Jul and Dec 2017 and compared with the testing sets. The fitting and predictive performance of these models was evaluated with mean absolute percentage error(MAPE),mean absolute error(MAE),root mean squared error(RMSE) and mean error rate(MER). Results The residuals of both α-Sutte and SutteSARIMA model fluctuated around 0,and the MAPE,MAE,RMSE and MER of α-Sutte, SARIMA and SutteSARIMA models were 7.68%,0.02,0.03,6.34%;12.10%,0.03,0.05,12.18%;and 7.72%,0.02,0.03,7.27%.The predicted MAPE,MAE,RMSE and MER were 17.13%,0.02,0.02,16.17%;15.32%,0.02,0.02,15.54%;and 5.88%,0.01,0.01,5.80%,respectively. SARIMA(1,1,3)(2,1,1)12was the optimal SARIMA model. Conclusion SutteSARIMA is the optimal model for prediction of hepatitis A incidence in China, which may be promising for prediction of other diseases.