Case study · 03
SkillMap AI
Student Support Prediction & Learning Analytics
SkillMap AI turns large-scale student interaction data into support predictions and teacher-facing recommendations through leakage-safe historical features, LightGBM modeling, and threshold tuning.
- Machine Learning
- Learning Analytics
- LightGBM
Problem
Large educational interaction datasets can reveal where students and skills may need support, but useful predictions require careful feature design, evaluation, and an interface that makes results actionable for educators.
Methodology
The pipeline uses historical features designed to avoid target leakage, trains a LightGBM model, and tunes the classification threshold to improve identification of support-needed cases.
- Approximately 525,000 raw interaction records
- Leakage-safe historical features
- LightGBM classification
- Threshold tuning
- Teacher-facing recommendations
Evaluation
The model reached 87.34% ROC-AUC. At the tuned threshold, support-needed recall reached 88.40% and support-needed F1 reached 78.63%.
Decision support
The dashboard is designed to help educators identify students and skills that may benefit from targeted support. Its outputs are recommendations for human review, not automated educational decisions.
Related research
Machine Learning · Learning Analytics · AI Evaluation