Predicting Dropout on MOOCs via Characterizing Personalized Learning Ability and Course Difficulty

Due to their extensive resources and high flexibility, Massive Open Online Courses (MOOCs) have emerged as a popular way of online learning for millions of people across the world. However, the high dropout rate has been identified as one of the critical issues related to many current MOOCs platforms. While numerous approaches have been suggested for predicting dropouts, most of existing methods simply focus on analyzing learners’ behavioral information, which ignore the impact of learner-specific learning ability and course difficulty, leading to the undesirable predictive performance. In this paper, we propose a novel deep learning-based model for dropout prediction, in which a multi-task learning structure is created to optimize the utilization of students’ learning strengths and course challenges for hidden representation learning. More specifically, predicting students’ learning ability and predicting course difficulty are modeled as two subsidiary subtasks within the framework of multi-task learning. Consequently, the supportive information is fully used for learning high-quality representations for the task of dropout prediction. Comprehensive experiments are conducted on the real-world dataset KDD CUP 2015, these findings highlight the efficacy of the suggested approach.