Predicting At-Risk Students' Academic Performance in an Online Learning Environment Using Learning Management System Interaction Data and the Random Forest Algorithm

Abstract

Purpose – This study aims to develop a model to identify at-risk students from LMS interaction data, analyzing how existing machine learning models can improve this identification.

Method – A machine learning model was created using five classifiers: random forest (RF), support vector machine, Naive Bayes, logistic regression, and K-nearest neighbor, to predict student performance from LMS interactions in a dataset of 486 students from a local university.

Results – The Random Forest algorithm achieved an MCC score of 66.42%, a Kappa score of 64.94%, and an F1 score of 66.62%.

Conclusion – LMS has enhanced education by improving accessibility and centralizing information, but challenges remain in identifying at-risk students. ML models like Random Forest show promise in addressing this issue.

Recommendations – Use more reliable datasets, explore imbalance treatment techniques, and integrate Random Forest predictive modeling to identify at-risk students in LMS.

Research Implications – This research seeks to promote the use of robust methods for improving predictive modeling accuracy using Random Forest to identify at-risk students.

Practical Implications – This research provides insights on predictive modeling using Random Forest and student interaction data from LMS to enable timely interventions and improve student success and learning outcomes.