Knowledge transfer by discriminative pre-training for academic performance prediction

Education is an area where data is scarce due to belated digital transformation or the high cost of acquiring data such as test scores. The label-scarcity problem brings a big challenge in taking machine learning approaches for academic performance prediction in intelligent tutoring systems(ITS).

Definition:

An intelligent tutoring system(ITS) is a computer-aided learning environment designed to provide immediate and customized instructions that are adaptive to a learner’s knowledge.

To this end, we propose DPA, a transfer learning framework with Discriminative Pre-training tasks for academic performance prediction. Experimental results show that DPA is a robust methodology impervious to the label scarcity problem.