STUDENT DROPOUT PREDICTION IN ONLINE COURSES BASED ON SUPERVISED CONTRASTIVE LEARNING

Abstract

Online learning provides numerous benefits and has become increasingly popular in recent years. However, the dropout rate in these learning environments is very high and thus represents a major issue for educational institutions. Early identification of students who are likely to drop out enables educators or course providers to offer appropriate support and improve student engagement. Some existing approaches for predicting student dropouts are based on traditional machine learning or deep learning techniques. However, unique characteristics of Massive Open Online Courses (MOOC) data pose significant challenges for effective dropout prediction. This work proposes a deep learning based method, called DP-SCL, for the student dropout prediction. The proposed method applies a supervised contrastive learning method in which a shared encoder with Long Short-Term Memory and Multi-Head Attention layers is trained using supervised contrastive loss. Experimental results on a real dataset demonstrate the strength of the proposed model compared with twelve other baseline methods. The source code of DP-SCL can be directly downloaded from our github page https://github.com/doanphong1995/DP-SCL