Data-Driven Stratification
Applies k-means clustering on 15 baseline variables (baseline EF, reaction time, anxiety, social preference, etc.). Identifies 4-5 distinct learner phenotypes.
Rapid Learners (Type A)
High baseline working memory, fast response time, low anxiety
✓ Benefit from accelerated progression
Careful Processors (Type B)
Slower response, high accuracy, high perfectionism
✓ Need confidence-building, avoid speed pressure
Social Learners (Type C)
High peer sensitivity, improve with competition
✓ Multiplayer/social features boost engagement
Stability-Seekers (Type D)
Prefer predictable schedules, sensitive to changes
✓ Consistent routine improves compliance
ML Pipeline
- • PCA dimensionality reduction (15 variables → 5 components, 82% variance explained)
- • Elbow method determines optimal k=4 clusters
- • Silhouette score validates cohesion (avg=0.68, good)
- • Cluster characteristics interpreted via domain expertise
Intervention Personalization
Type A gets adaptive difficulty acceleration; Type B gets accuracy feedback; Type C gets tournament modes; Type D gets fixed weekly schedule. Average ROI increase: 31%.
