Clustering Techniques to Identify Low-Engagement Student Levels

Dropout and failure rates are a major challenge with online learning. Virtual Learning Environments (VLE) as used in universities have difficulty in monitoring student engagement during the courses with increased rates of students dropping out. The aim of this research is to develop a data-driven clustering model aimed at identifying low student engagement during the early stages of the course cycle. This approach, is used to demonstrate how cluster analysis can be used to group the students who are having similar online behaviour patterns in the VLEs. A freely accessible Open University Learning Analytics (OULA) dataset that consists of more than 30,000 students and 7 courses is used to build clustering model based on a set of unique features, extracted from the student’s engagement platform and academic performance. This research has been carried out using three unsupervised clustering algorithms, namely Gaussian Mixture, Hierarchical and K-prototype. Models efficiency is measured using a clustering evaluation metric to find the best fit model. Results demonstrate that the K-Prototype model clustered the low-engagement students more accurately than the other proposed models and generated highly partitioned clusters. This research can be used to help instructors monitor student online engagement and provide additional supports to reduce the dropout rate.