Objective: To explore the application of seasonal time series ARIMA model in prediction of malaria incidence in an unstable malaria area.
Methods: SPSS13.0 software was used to construct the ARIMA model based on the monthly malaria incidence of Huaiyuan and Tongbai counties in Huaihe River Valley, from Jan. 1998 to Dec. 2005, with consideration of residual un-correlation and conclusion. Akaike's information criterion (AIC) and Bayesian information criterion (BIC) were used to confirm the fitness of model. The constructed model was then applied to predict the monthly malaria incidence in 2006 and the incidence from ARIMA model was compared with the actual incidence, so as to evaluate the model's validity. Malaria incidence of 2007 was predicted by ARIMA model based on malaria incidence from 1998 to 2006.
Results: Statistics assisted estimation of the significance of the fitted autoregressive and seasonal moving average coefficients (AR1=0.512, SMA1=0.609, P<0.01). ARIMA (1,0,0)(0,1,1)12 model, with AIC=67.01, BIC= 71.87 and white noise for predicting error, exactly fitted the incidence of the previous monthly incidence from Jan. 1998 to Dec. 2005, and the predicted monthly incidence in 2006 by the model was consistent with the actual incidence. Malaria incidence of 2007 would be 106.50/100 000, with a peak incidence during July and October.
Conclusion: The model of ARIMA seems to be an appropriate model to fit exactly the changes of malaria incidence and to predict the future incidence trend, with a high prediction precision of short term time series.