Time series techniques proposed by Box-Jenkins were applied to forecast the number of Sexually Transmitted Diseases (STD) cases. This methodology is useful for forecasting with the use of model constructed, which is based solely on empirical past values.
The data used in this study comes from the STD Report at the Preventive medicine section of the 18th MEDCOM, USFK. It consists of the monthly notification of the STD cases who are available from 1985.
The Preventive Medicine section has been estimated that the data quality is of a particularly high standard since estabbshing the report system. Which is due mainly to military status of the reporters, case contact interviewers. Furthemore, more than 90 percent of all STD cases have been succeed in completing the report system. This information source has been considered sufficiently informative to apply time-series analysis for short-term forecasting of STD cases. During the first 2 years after starting the surveillance system, we are not making sure for reporting rates, and so we excluded data for that period. The period analysed includes 120 months of observation from 1987 to December 1996.
The ultimate purpose of this study is to obtain a forecast function which is such that the mean square of the deviations ZtZt(I) between the actual and forcadted values is as much as possible for each lead time 1.
Two possible model were chosen: each one of seasonal and non-seasonal ARLMA model. The former model was fitted well for forecasting purpose. It corresponds to a multiplicative seasonal autoregressive integrated moving-average(ARIMA) model: ARIMA (0,1,1)x(0,1,1)12 for the original series Zt which can be written as:
Where Wt +AD0- (Zt - Zt )-(Zt-12-Zt-13), with a mean of 2.602. The modified Box-Pierce 2 statistic was 2193 (lag of 24: df,21:p +AD0- 0.403),indicating that model adequately described the data.
Separate models for different change-point period need to be tried for the comparison of forecasting errors. A Box-Jenkins type of modeling for STD data seems well adapted to forecast understandable results.
For forecast purpose. A periodic updating on the original series must be performed so as to get full advantage to the ARIMA model. Moreover multivariate time series analysis such as intervention analysis and transfer function noise model would be necessary in this kind of topic, which is essential procedure for more than two input series, in further work on issue, it would also be neccssary to include prior history of STDs as a new input series.
|