Case study · 04
SkillMap AI
Student Support Prediction & Learning Analytics
SkillMap AI turns large-scale learning-interaction data into practical support signals for educators. It combines historical feature engineering, machine-learning evaluation, and an interactive dashboard for reviewing students, skills, and recommended actions.
- Machine Learning
- Learning Analytics
- LightGBM
Problem
Large learning datasets can reveal where students and skills may need support, but useful predictions require careful modeling and a clear interface for educators.
Approach
The project starts with approximately 525,000 raw ASSISTments interaction records. It builds chronological historical features while avoiding same-row behavior that could leak the answer, compares candidate models, and selects and tunes LightGBM.
The decision threshold is tuned to prioritize finding more support-needed cases.
Results
The model reached 87.34% ROC-AUC. At a threshold of 0.40, support-needed recall reached 88.40% and support-needed F1 reached 78.63%.
Decision support
The dashboard helps educators inspect skills that may need review, students who may need follow-up, class-level patterns, and recommended actions.
It supports educator review and does not make automated educational decisions.
Related areas
Machine Learning · Learning Analytics · AI Evaluation