Upcoming Talks 

CASI Abstract - May Galway, 2025

Scalable AI-Based Framework for Predicting Learning Trajectories: ATU Case Study

Accurately understanding and predicting student learning trajectories is crucial to transforming educational outcomes, especially in the critical fields of mathematics and computing. Traditional learning analytics typically focus on structured engagement data, often overlooking factors from cognitive neuropsychology such as self-regulated learning, self-efficacy, and other metacognitive elements that significantly influence student success. This study presents a scalable, data-driven framework that integrates machine learning, natural language processing (NLP), and advanced statistical modeling to explore and predict learning trajectories at Atlantic Technological University (ATU). By incorporating these factors, the framework provides a more comprehensive understanding of student learning beyond traditional metrics.

Aligned with principles of educational neuroscience, this framework emphasizes learning progressions, which reflect how neural connections strengthen over time with practice, mirroring the brain’s natural process of learning. Data is continuously collected at various stages of student engagement, providing real-time insights into how learners develop mastery of specific competencies. The framework acknowledges that each student follows a unique learning trajectory shaped by cognitive abilities, prior knowledge, motivation, participation, environmental factors, and mental health. By monitoring key learning indicators (e.g., quiz efforts, sentiment analysis, and participation patterns) at strategic points, the framework enables the identification of students who may benefit from additional support, personalized feedback and educational interventions. The model combines temporal dynamics derived from time-series sampling of VLE engagement metrics and self-efficacy insights from validated instruments New Generalized Self-Efficacy Scale (NGSE) and open-ended Self-Regulated Learning (SRL) responses.

A critical component of this framework is the collaboration between ATU’s computing services and the learning analytics research team, ensuring a robust infrastructure for real-time data processing and secure, ethical data handling. The framework leverages a Lakehouse architecture within a Fabric environment, enabling seamless integration of predictive models while maintaining stringent data privacy and security standards. The framework streamlines several processes, including data cleaning and transformation, handling missing values, normalizing features, and resampling datasets to address class imbalances. Predictive models such as Random Forest and XGBoost are used to forecast academic performance, while Transformer-based architectures and BERTopic analyze learners language patterns and sentiment. Survival analysis and anomaly detection techniques highlight periods when students may need additional support based on their engagement behaviours, providing proactive, data-driven insights. Model interpretability is enhanced through SHAP and permutation importance, ensuring actionable insights for educators.

Data analysis from Year 1 and Year 4 Computing and Science students reveals that consistent effort through mastery quizzes and positive sentiment trends strongly correlate with academic progression. Predictive models achieved 89% recall and 72% accuracy for identifying at-risk student groups by Week 5. These findings demonstrate AI’s potential to shift learning analytics from reactive assessments to proactive, data-driven support, ultimately enhancing student success. The research contributes to optimizing student success in higher education by offering a scalable, statistically rigorous framework for personalized educational strategies. This approach’s practical implications extend beyond ATU, offering a model that can be adapted to other institutions with potential applications across higher education.