Projects using scikit-learn in SurabayaProjects using scikit-learn in SurabayaAI Student Success Intelligence Platform
A twelve‑module analytics platform analyzed 50,000 learners across six countries to predict dropout, model engagement, and simulate interventions, finding that a composite Student Engagement Index (SEI)—built from Time Commitment, Academic Quality, Platform Activity, and Social Learning—is the strongest predictor of dropout (behavior beats demographics), an ensemble of XGBoost/LightGBM/CatBoost achieved 99.72% AUC and F1 = 0.9522, risk tiers were highly precise (Low Risk = 0.0% dropout; Critical Risk = 99.7%), multi‑dimensional “Full Interventions” produced the largest simulated risk reductions, and correcting a data‑leakage issue (attendance proxy) was essential to preserve model integrity. Global Retail Intelligence System: Product Success Prediction and Strategic Market Analysis
A multi-stage ML pipeline analyzed 44,888 Adidas SKUs using XGBoost and Random Forest to predict product success, demand trajectories, and stockout risk, finding that subcategory is the dominant success driver (~6× more explanatory than price, discount, or geography), the Success Classifier reached 94.3% accuracy and the Stockout Risk model 0.99 ROC‑AUC, 42.5% of products carry markdowns with deep discounts (≥30%) often eroding margins, 323 high-performing SKUs are under‑distributed and present near‑term expansion opportunities, the Budget tier outperforms Premium/Luxury in conversion to high performers, and 653 SKUs were flagged as high demand with elevated stockout risk requiring urgent replenishment. 
