Classifying MOOC Students Using k-Nearest Neighbors and Decision Trees for Early Detection of At-Risk Learners

Abstract

Massive Open Online Courses (MOOCs) have expanded access to education, but they consistently face issues with a high dropout rate. At-risk students need to be identified early so that they can be promptly intervened upon and achieve better academic results. This study investigates the use of supervised machine learning, specifically Decision Trees (DT) and k-Nearest Neighbors (kNN), to classify MOOC students from Institution A based on behavioural and engagement indicators. The predictive features include the number of comments posted in discussion forums, the number of kudos received from other learners, overall course progress, and time spent on learning pages. These variables represent both social interaction and individual engagement dimensions of learning behaviour. A 552 dataset of MOOC participants was pre-processed and analysed, followed by classification experiments using DT and kNN. Model performance was evaluated using accuracy, precision, recall, and F1-score. Evaluation results showed that the Decision Tree algorithm delivered the best performance, achieving a perfect score of 98.19% accuracy, 98.23% precision, 98.19% recall, and a 98.12% F1-score. Meanwhile, the KNN algorithm achieved 98.19% accuracy, 98.23% precision, 98.19% recall, and a 98.12% F1 Score. These findings suggest that the Decision Tree model is more effective for classifying customer loyalty and can be utilised as a decision support tool in identifying potential at-risk learners.