Mining Sequential Learning Trajectories With Hidden Markov Models For Early Prediction of At-Risk Students in E-Learning Environments

With the onset of online education via technology-enhanced learning platforms, large amount of educational data is being generated in the form of logs, clickstreams, performance, etc. These Virtual Learning Environments provide an opportunity to the researchers for the application of educational data mining and learning analytics, for mining the students learning behavior. This further helps them in data-driven decision making through timely intervention via early warning systems (EWS), reflecting and optimizing educational environments, and refining pedagogical designs. In this, the role of EWS is to timely identify the at-risk students. This study proposes a modeling methodology deploying interpretable Hidden Markov Model for mining of the sequential learning behavior built upon derived performance features from light-weight assessments. The public OULA dataset having diversified courses and 32 593 student records is used for validation. The results on the unseen test data achieve a classification accuracy ranging from 87.67% to 94.83% and AUC from 0.927 to 0.989, and outperforms other baseline models. For implementation of EWS, the study also predicts the optimal time-period, during the first and second quarter of the course with sufficient number of light-weight assessments in place. With the outcomes, this study tries to establish an efficient generalized modeling framework that may lead the higher educational institutes toward sustainable development.