Prediction anf Optimization for World University Ranking of Prince of Songkla University

Abstract

World university rankings reflect quality of higher educational institutions. Therefore, there are high competitions among institutions to be in a higher rank. For that reason, most institutions are discussing about how to increase their global rank. Two subprojects were investigated in this thesis. The first project was to forecast a possible tendency of indicator scores of the Times Higher Education (THE) and the Quacquarelli Symonds Rankings (QS) of Prince of Songkla University (PSU), Thailand. The exponential smoothing, the moving average and the ARIMA techniques have been compared to find out which technique was more appropriate to predict the performance of PSU based on the indicators of the THE and QS. The data such as the number of academic staff, the number of full-time students, etc., from 1994 to 2014 have been used in the analysis. The data were firstly classified into two classes relying on trend and no existing trend. For no-trend series, the single moving average (SMA) and the single exponential smoothing (SES) were chosen to predict the data tendency. For the other class, the double moving average (DMA) and Holt's method were applied. In addition, ARIMA was also used to forecast for both groups. According to the mean squared error (MSE), the SES is the most appropriate technique for the no-trend series, whereas the Holt's method is suitable for the trend series. The second project was to find optimal values for each of our studied indicators: faculty students ratio, citations per faculty, proportion of international faculty, and proportion of international students, that maximizes the overall score of QS Ranking. Those four indicators are commonly used in most university ranking systems and considered to be controllable. An approach of optimization using maximization of nonlinear programming problem in which the objective function was constructed from normalization and weighting was applied throughout this research. Three cases of constraints that are different in boundary determination were considered. The results from the analysis showed that the optimal values were varied depending on the constraints. The final decision for the optimal values is based on context, ability and policy of an individual educational institution.