To evaluate the performance of machine learning algorithms in predicting student dropout on MOOC platforms

Abstract

Online learning using Massive Open Online courses(MOOCs) has gained a lot of hype in recent years due to its great potential in having the widest reach in delivering the state-of-the-art resources to the unlimited number of online learners without limiting itself to any geographical boundary. Along with gaining popularity, MOOCs have been facing challenges like high attrition or dropout rate since its birth. The main motivating factor behind the study is to fill the gap which has been there because of very limited literature available there to find the real cause behind these challenges. The current study tries to find the solution of the said challenges by finding the significant contributing factors which highly affect the target variable in the study which is number of certified students in this case. The dataset used in this paper is publicly available in dataverse repository of Harvard university. The dataset is a compilation of student clickstream log data consisting of 641138 instances of enrolled students in various MOOC courses of Harvard and MIT. The study evaluates machine learning models like logistic regression, decision tree, random forest, K-Nearest Neighbor to determine their efficiency in predicting the student dropout. The results of this study can be used to create a framework for recommending necessary actions to the at-risk students to reduce the dropout rate.