An empirical comparison of models for dropout prophecy in MOOCs

Abstract

MOOCs are Massive Open Online Courses, which are offered on web and have become a focal point for students preferring e-learning. Regardless of enormous enrollment of students in MOOCs, the amount of dropout students in these courses are too high. For the success of MOOCs, their dropout rates must decrease. As the proportion of continuing and dropout students in MOOCs varies considerably, the class imbalance problem has been observed in normally all MOOCs dataset. Researchers have developed models to predict the dropout students in MOOCs using different techniques. The features, which affect these models, can be obtained during registration and interaction of students with MOOCs' portal. Using results of these models, appropriate actions can be taken for students in order to retain them. In this paper, we have created four models using various machine learning techniques over publically available dataset. After the empirical analysis and evaluation of these models, we found that model created by Naïve Bayes technique performed well for imbalance class data of MOOCs.