Combining transfer learning and co-training for student classification in an academic credit system

Abstract

Student classification is one of the popular educational data mining tasks to early predict in-trouble students in an educational system for appropriate and timely support. Besides, an academic credit system is nowadays widely-used all over the world due to its flexibility in order that teaching and learning activities can be efficiently conducted. Nevertheless, its flexibility might lead to the heterogeneity in educational data gathered in the system over time. Also, the changes in one educational program or the differences between the educational programs might hinder the exploitation of knowledge discovered in educational data. Indeed, a conventional student classification model built on one program can not be utilized for other programs. In this paper, our work proposes a novel approach to student classification in an academic credit system by combining transfer learning and co-training. A resulting model can predict a study status of a student enrolled in one educational program effectively by using a classification model enhanced by transfer learning and co-training techniques on educational data from another program. In addition, our approach can deal with the sparseness in educational data sets for early in-trouble student prediction. Experimental results on real data sets have shown that our approach could provide an effective solution to student classification in an academic credit system.