Abstract:

Background:
Prospective students apply to the Singapore University of Social Sciences (SUSS) and other Universities in Singapore towards the end of the calendar year. SUSS admissions staff will shortlist applicants for Full Time (FT) programmes based on their past academic performance, portfolio showcasing non-academic achievements and a reflection essay. Shortlisted students will be invited to come for a selection session at SUSS. They will write an essay, do a cognitive exercise, participate in a group discussion and have an individual or cluster interview. Students will be offered places in SUSS based on their past academic performance, portfolio work and performance in the selection tests and interviews.
This paper discusses the prediction of applicants’ academic performance based on information (i.e., variables) available at the point of application to aid in the shortlist and offer decisions. In view of limited places on offer, the University wants to give them to applicants with a reasonable chance of success.

Approach:
The majority of applicants are from Singapore’s five polytechnics and the rest are from junior colleges, private institutions and schools offering the International Baccalaureate (IB) diploma (Li, 2019). Data of past applicants including their demographics, prior education (e.g. type of qualification, institution) and prior academic performance at the point of application (including O, A level and polytechnic results) are gathered. These are combined with derived variables to train models to predict an applicant’s Cumulative Grade Point Average (CGPA) as a proxy for potential academic performance.
Statistical (e.g. stepwise regression) and Machine Learning (e.g. decision tree) data mining models are trained to predict the CGPA at the time of shortlisting. Selection scores including test, interview, portfolio and essay scores are added to the dataset and various models are again trained to predict the CGPA at the time of offer. The trained models are compared using various metrics and the champion Shortlist and Offer models are identified. The champion models are then used to score the applicant’s data resulting in each applicant having a predicted shortlist and offer CGPA.

Deployment:
Dashboards are used as a graphical user interface that provides interactive views into key measures and trends to assist decision makers in SUSS (Tang, 2019). Dashboards are also built to deploy the predicted results allowing decision makers to view them along with other data collected on the applicants. They allow users to perform guided analysis and zoom in on individual applicants. In addition to the usual information used by SUSS faculty and staff to make the Shortlist and Offer decisions, the prediction model provides users with additional insight into potential academic performance of the applicants. It is hoped that the decision makers will carefully consider this along with the other available information and use their domain and prior knowledge to make a more informed decision to shortlist and offer a place for students in the University.

Keywords:

Academic performance, prediction, data mining, dashboards.