Predictive Analysis of Student Engagement in Adaptive LMS Platforms using Time-Series Modeling

Student engagement is a critical determinant of learning effectiveness in digital education environments, yet most Learning Management Systems (LMS) remain reactive, relying on descriptive analytics that fail to anticipate disengagement in a timely manner. This study proposes and evaluates a time-series–based predictive framework for modeling student engagement and integrating engagement forecasts into adaptive LMS mechanisms. Engagement is operationalized as a composite temporal signal derived from fine-grained LMS interaction logs, including session duration, content access patterns, assessment activity, and adaptive system interactions, aggregated into fixed time windows to preserve sequential dependency. Empirical analysis demonstrates that student engagement exhibits strong temporal continuity and structured oscillatory patterns, validating the suitability of time-series modeling. Rolling-window forecasting experiments show consistently low prediction error across multiple temporal segments, with Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) remaining stable under varying engagement volatility. When engagement predictions are embedded into adaptive intervention policies, the adaptive LMS achieves a measurable increase in mean engagement levels, a reduction in engagement variance, and a significant improvement in recovery speed following low-engagement episodes compared to non-adaptive LMS configurations. Longitudinal comparison further reveals that predictive adaptive systems maintain higher minimum engagement thresholds and reduce the frequency of sustained disengagement states. These findings indicate that engagement prediction functions effectively as a system-level control signal, transforming adaptive LMS platforms from reactive content delivery systems into proactive engagement management environments. The study contributes an engineering-oriented perspective by framing adaptive learning as a closed-loop system that continuously senses, predicts, and responds to learner behavior. The results provide empirical evidence that even lightweight time-series models, when tightly integrated with adaptive decision mechanisms, can substantially enhance engagement stability and learning continuity in large-scale LMS deployments.