Evaluating Models for a Higher Education Course Recommender System Using State Exam Results

Abstract

When young people approach the end of their schooling, they are faced with a plethora of often daunting decisions, including whether to go to University or other further education institution, and what further education course is most suitable for them. In this paper, we propose using result data obtained from the Colombian Saber 11/T&T/Pro state exams as input for a higher education course recommender system. We compare five different recommender models by analyzing precision, recall, ROC curves and prediction error. Our findings are that user-based collaborative filtering, and the model that recommends the most popular courses, are the ones that perform best. We note that while the context of this work is in Colombia, most other countries have similar or equivalent state exams. It can therefore be expected that our research findings can be more generally applied to other contexts. As further work, we hope to deploy this recommender system as a mobile telephone application for young people to use to help them choose higher education courses.