Abstract:

Understanding learner engagement in digital learning environments remains a major challenge in educational research. While learning management systems generate extensive behavioural data, most learning analytics studies rely on static indicators such as click counts or time-on-task, which fail to capture the dynamic nature of learner engagement. In practice, engagement and cognitive effort fluctuate over time as students interact with course materials, reflecting changes in attention, persistence, and cognitive energy during learning activities.

This study proposes a machine learning framework for modelling latent cognitive energy dynamics using behavioural interaction data from online learning environments. Using the Open University Learning Analytics Dataset (OULAD), which contains behavioural traces from more than 30,000 students, temporal engagement features were derived from activity logs in the virtual learning environment. A Hidden Markov Model (HMM) was applied to infer latent behavioural states representing different engagement regimes over time. The resulting state sequences were analysed to examine engagement dynamics and their relationship to academic outcomes.

Results indicate that student behaviour evolves through three persistent engagement states, reflecting different levels of cognitive energy during learning. A Random Forest regression model incorporating cognitive state features achieved R² = 0.134, outperforming a baseline model based solely on traditional engagement metrics. Clustering analysis further revealed four learner cognitive energy archetypes, and statistical testing confirmed significant differences in performance across these groups (F = 73.73, p < .001).

These findings demonstrate that modelling temporal behavioural dynamics provides deeper insight into learner engagement and supports the development of more adaptive, data-driven educational technologies. By integrating probabilistic sequence modelling, predictive analytics, and behavioural clustering, this study demonstrates how machine learning can uncover latent engagement structures from large-scale educational data. The proposed framework contributes to the development of human-centred artificial intelligence in education, offering new opportunities for identifying engagement fluctuations, understanding learner behaviour, and designing adaptive learning systems capable of supporting students more effectively.

Keywords:

Learning Analytics, Artificial Intelligence in Education, Machine Learning, Hidden Markov Models, Student and Cognitive Engagement, Behavioral Data Mining, Online Learning, Educational Data Mining.