Predicting student success using student engagement in the online component of a blended-learning course

Abstract

There has been a surge in student failure rates in blended-learning courses in recent times, which has generated considerable research interests. Engagement is identified as one of the core metrics for measuring students’ success or failure in any learning system. This study utilizes machine learning algorithms on students’ log-file data col lected from an LMS to predict student success and increase their throughput rates. The machine learning predictive models considered in this study are Logistic Re gression, Support Vector Machines, Naïve Bayes, Decision Tree, Random Forests, Gra dient Boosting Tree, Multilayer Perceptron Neural Network, and Linear Discriminant Analysis. The study presents the advantage of using SMOTE sampling in handling imbalance class problems over Random Under-Sampling and Random Over-Sampling Techniques. The Random Forests performance surpassed the other machine learning models in this study with an accuracy value of 91%, AUC of 0.90, and F1-score of 0.98. The results provide an automatic predictive model for timely identification of learners at risk of failing in their courses for instructor early intervention. The significance of this study is to provide a feedback tool on engagement for an increase in student performance.