摘要
目的利用Prophet模型、差分自回归移动平均模型(autoregressive integrated moving average model,ARIMA)和NeuralProphet模型探索我国2008—2018年手足口病(hand-foot-and-mouth disease,HFMD)发病趋势和季节性特征,比较3个模型的拟合及预测效果。方法收集我国2008—2018年手足口病月发病率数据,2008年1月—2017年12月的发病数据用于构建Prophet、ARIMA和NeuralProphet 3种预测模型,2018年1月—12月的发病数据作为验证集,分析评价各模型的预测性能,选取均方根误差(root mean square error,RMSE)、平均绝对误差(mean absolute error,MAE)作为指标评价预测模型的拟合及预测效果。结果我国手足口病存在2个发病高峰,主高峰在4~7月,次高峰在9~11月;春节的假期效应远高于其他假期;Prophet模型为最优预测模型。Prophet模型、SARIMA(0,1,3)×(0,1,1)_(12)模型和NeuralProphet模型预测的均方根误差(root mean squareerror,RMSE)、平均绝对误差(mean absolute error,MAE)依次为4.09、3.17,4.26、3.61,4.25、3.79。结论Prophet模型的拟合预测精度优于ARIMA模型和NeuralProphet模型,能很好地拟合我国手足口发病趋势,具有较好的应用前景,可为手足口病的防控工作提供决策性依据。
Objective To explore the incidence trend and seasonal characteristics of hand-foot-and-mouth disease(HFMD)in China from 2008 to 2018 by using Prophet model,autoregressive integrated moving average model(ARIMA)and NeuralProphet model,and to compare the fitting and prediction effects of the three models.Methods The monthly incidence data of HFMD from 2008 to 2018 in China were collected,among which the incidence data from January 2008 to December 2017 were used to develop three prediction models of Prophet,ARIMA and NeuralProphet,and the incidence data from January to December 2018 were used as a validation set.The prediction performance of each model was analyzed and evaluated.Root mean square error(RMSE)and mean absolute error(MAE)were selected as indexes to evaluate the fitting and prediction effect of the prediction model.Results There were two peaks of HFMD in China,the main peak appeared from April to July,and the secondary peak appeared from September to November;the holiday effect of Spring Festival was much higher than that of other holidays;prophet model had the best fitting and prediction performance,the evaluation indexes RMSE and MAE were 4.09,3.17 for Prophet model,4.26,3.61 for SARIMA(0,1,3)×(0,1,1)_(12) model,and 4.25,3.79 for NeuralProphet model,respectively.Conclusion The ftting and forecasting precision of Prophet model has high accuracy,better than ARIMA model and NeuralProphet model,which can well fit the trend of HFMD in China.With good application value,the Prophet model could provide decision-making basis for the prevention and control of HFMD.
作者
周海涛
王玥
岳婷雨
陈威
胡斌
ZHOU Haitao;WANG Yue;YUE Tingyu;CHEN Wei;HU Bin(Department of Public Health,Xuzhou Medical University School,Xuzhou,Jiangsu 221004,China)
出处
《医学动物防制》
2023年第8期715-721,共7页
Journal of Medical Pest Control
基金
江苏省研究生科研与实践创新项目(SJCX23_1407)
江苏省研究生科研与实践创新项目(JCX22_1291)
徐州市科技计划项目(KC20200)。
关键词
手足口病
Prophet模型
季节ARIMA模型
时间序列分析
Hand-foot-and-mouth disease
Prophet model
Seasonal autoregressive integrated moving average model
Time series analvsis
作者简介
周海涛(1995-),男,在读硕士研究生,研究方向:公共卫生传染病预警;通信作者:胡斌,E-mail:hubin@xzhmu.edu.cn。
