A Machine Learning Approach to Predict Learner Dropout in Online Courses

Abstract

While the development of Massive Open Online Courses (MOOCs) and other online learning systems have changed
the way education is delivered, dropout rates continue to be a challenge. This study suggests a behavioral and
participation pattern predictive model for dropout using machine learning algorithms. The analysis focuses on a dataset
from a popular MOOC, which has key features like clickstream data, forum participation, assignment submissions,
and time spent on course videos. Supervised learning algorithms such as Random Forest, Logistic Regression, Support
Vector Machine (SVM), and Gradient Boosting were trained and evaluated. The highest performance was from the
Random Forest classifier with an F1 of 0.89 and AUC of 0.93. This study also evaluates the feature sets and models
to find the most accurate predictors of dropout and measures the predictive power of diversified feature sets. These
results demonstrate that there is opportunity to apply machine learning techniques to unstructured data in order to flag
students likely to drop out prior to course completion, enabling timely assistance which has demonstrated redemption
potential for retention. The increasing amount of available data regarding student interaction with online learning
environments fueled developments in educational data mining, which the study now adds to along with practical value
for educators, administrators, and developers striving to enhance learner achievement in digital educational settings.