Discussion on selection of time series model for prediction infectious diseases
Objective To compare the fitting and forecasting performance of different time series models for different notifiable infectious diseases,so as to provide reference for the selection of forecasting and early warning models.Methods Weekly incidence data of tuberculosis(TB),hepatitis C(HCV),hemorrhagic fever with renal syndrome(HFRS),hand-foot-mouth disease(HFMD),and other infectious diarrhea(Diarrhea)in Jingzhou City from 2012 to 2018 were used as examples to establish 8 time series models,including SARIMA,ETS,TBATS,NNETAR,SPLINE,THETA,prophet and BSTS.The reported incidence of 5 types of notifiable infectious diseases in the 1-52 weeks of 2019 were predicted and compared with the actual values.Mean Absolute Percentage Error(MAPE)were used to evaluate the fitting and prediction performance of different models.Results The best fitting model for Diarrhea(9.12％)and HFMD(13.80％)were SPLINE.NNETAR model showed the best fitting performance for TB(3.82％),HCV(15.53％),and HFRS(1.83％).In terms of Diarrhea(20.27％)and HFRS(34.73％),TBATS was the best predictive model.The best predictive model for HFMD(65.67％)was prophet.The SARIMA model had the best predictive performance for TB(16.66％)and HCV(26.19％).As for Diarrhea(14.89％),the model with the highest comprehensive accuracy of fitting and prediction was TBATS.The model TBATS for HFMD(32.05％),TB(6.52％),HCV(19.92％),and HFRS(8.50％)had the highest comprehensive accuracy of fitting and predicting.Conclusions Different models have different fitting and prediction accuracy for different types of infectious diseases,and the best one should be selected according to research needs.
predictionfittinginfectious diseasetime series model
万方数据
维普
