A comparison of errors in forecasting educational time series data with nonstationary between box-jenkins techniques using structural equation model and leading indicator
Abstract:
This research aimed at comparison of errors between Box-Jenkins using structural equation model method and Box-Jenkins using leading indicator method in forecasting educational time series data with nonstationary in 5 lead times forecasting, and at checking the forecasting results using 6 corrected measures of forecasting; RMSE, MAPE, MdAPE, MdRAE, GMFtAE and Percent Better as criteria. Data base used in this study was Bumpen Pidchids, and the researcher collected the more up to date data from the Educational Information Center, Faculty of Education, Chulalongkorn University. The research instrument was data recording form. The data were analyzed using graph and regression analysis to check for secular trends and seasonal variations, then applying Box-Jenkins using structural equation model, Box-Jenkins using leading indicator, and Box-Jenkins methods to forecast and check for 6 errors. The research findings were as follows : 1. General books in circulation showed quadratic trend and seasonal variation at .05 significance level, and the integrated model was additive. Reserved books in circulation showed linear trend and seasonal variation at .01 significance level, and the integrated model was additive. Theses in circulation and visitors showed quadratic trend and seasonal variation at .05 significance level, and the integrated model was multiplicative. 2. The results of three data sets showed that the appropriate model for general books in circulation was IMA(2,1)xSARI(1,1)12, and forecasted values in December 2542 to April 2543 would be 7,105, 8,020, 8,903, 4,906 and 5,364, respectively. The appropriate model for reserved books in circulation was ARI(1,1)xSARI(1,1)l2, and forecasted values in June 2539 to October 2539 would be 169, 335, 410, 264 and 162, respectively. The appropriate model for theses in circulation was ARI(2,2)xSARI(1,1)12, and forecast values in December 2542 to April 2543 would be 8,871, 9,445, 9,989, 10,281 and 9,240, respectively. 3. The forecasting results from Box-Jenkins method used in forecasting general books and reserved books in circulation and the Box-Jenkins using structural equation model method used เท forecasting theses in circulation had the least error.