Application Of Time Series Analysis In Forecasting Measles Incidence In Gansu Province

Objective: Measles is a serious health hazard disease for children's respiratory; the national Ministry of Health issued the "2006-2012 National Plan of Action for the elimination of measles". Request clearly that the incidence of measles must be control under the 1 / 1,000,000 in 2012; the measles elimination becomes another major challenge for immunization planning of Gansu province. Therefore, in this issue, trying to establish the statistical prediction model of measles incidence of Gansu and to forecast the trend of its incidence; Early prediction can provide a scientific basis for development of prevention and control of measles in the recent or long-term coping strategies in Gansu Province, and further achieve the incidence of measles control in the 1 /1000000 of the National Action Plan of the measles elimination in 2012Methods: collected measles surveillance data of Gansu CDC from 1995 to 2007, With different time series forecasting model, including index method, the moving-average method, exponential smoothing method and mixed autoregressive moving average model, establish Corresponding prediction model, to fit and prediction its incidence. Comparing the fitting and prediction accuracy of models, discussed analysis results combination with the Measles control actual work, and identify the better model for measles finally.Results: establishment of 3 kids of time series model1 Random time series of mixed autoregressive moving average model (ARIMA)Monthly predict model ARIMA (1,0,0) (0,1,1)₁₂ and Quarterly predict model ARIMA (1,0,0) (1,1,0)₄, that is:Monthly predict model: (1 - B) (1 - B12 ) Zt = (1 - 0.685B) (1- 0.955B12) atQuarterly predict model: (1 - B) (1 - B4) Zt = (1 - 0. 539B) (1 + 0.541B4) at2 Deterministic time series of trend modelThis study also use moving-average models and exponential smoothing model to establish predicted model of monthly and quarterly measles morbidity, that is. Moving-average models:Monthly predict model: Xi=(0.389－0.0261×i)×1.0463×LiQuarterly predict model: Xi = (2.35-0.5×i)×1.003×LiExponential smoothing modelMonthly predict model:α=0.65,γ=0.00,β=1.0, SSE=32.30 (d f=95)。Quarterly predict model:α=0.3,γ=0.00,β=0.8, SSE=129.80 (d f=31)。3 Deterministic time series of index methodUsing the average growth rate method and the average speed of development method to forecast the incidence of monthly and quarterly, according to the average growth rate method , the average growth rate of incidence in 1998 - 2006 are 0.0038/100,000 (level method)å’Œ0.433/100,000 (Summation method ) ,so the predicted incidence of 2007 are 6.394/100,000 (level method) and 10.257/100,000 (Summation method ) ;according to the average speed of development method , the average speed of development of incidence in 1998 - 2006 are 100.06% (geometry method) and 110.16% (equation method ) ,so the predicted incidence of 2007 are 6.394/100,000 (geometry method) and15.194/100,000 (equation method ) .Conclusion: Through comparison among those models, comprehensive analysis combined with the epidemiological characteristics of measles and the actual work situation of against measles, and concludes that the average growth rate and the average speed of development is not suitable to predict the incidence of measles. Moving average method, exponential smoothing and ARIMA model are all can be used to predict the future incidence of measles, but ARIMA model is better than others.In the specific application process, as traditional forecasting methods, Moving average and exponential smoothing method are simple to use; ARIMA model is a method of establish modeling which was proposed for sequence of time-sensitive, it analyses the sequence value of time point of each seasonal cycle and extracts Seasonal trend; while extracts non-seasonal components according to the internal sequence value changes in each Seasonal cycle. Thus the moving-average method, exponential smoothing and ARIMA model can be used in conjunction, and supplied each other. Comprehensively speaking, the program of solve problem is not unique in practical work, but there should have a better solution. When making disease incidence forecast, a better program should be selected based on compare the advantages and disadvantages of various options scheme under the limited condition, establish the best model finally.