Determining patterns of student graduation using a bi-level learning framework

Abstract

The practice of data science, artificial intelligence (AI) in general, has expanded greatly in terms of both theoretical and application domains. Many existing and new problems have been tackled using different reasoning and learning methods. These include the research subject, generally referred to as education data mining (or EDM). Among many issues that have been studied in this EMD community, student performance and achievement provide an interesting, yet useful result to shaping effective learning style and academic consultation. Specific to this work at Mae Fah Luang University, the pattern of students’ graduation is determined based on their profile of performance in different categories of courses. This course-group approach is picked up to generalize the framework for various undergraduation programmes. In that, a bi-level learning method is proposed in order to predict the length of study before graduation. At the first tier, clustering is applied to derive major types of performance profiles, for which classification models can be developed to refine the prediction further. With the experiments on a real data collection, this framework usually provides accurate predictive outcomes, using several conventional classification techniques.