Student Early Warning System

Architected a multi-scenario Machine Learning pipeline to classify university dropout risks across 7 academic semesters, achieving up to 81.3% peak accuracy and 75.0% recall using XGBoost and Random Forest.

Engineered an automated data pipeline to process 3,000+ raw records and extract 7 time-series predictive indicators, optimizing an 80:20 imbalanced dataset via strategic threshold calibration (0.42).

Deployed the inference engine as an interactive web application using Streamlit, enabling dynamic state management and sub-second evaluation across 7 dynamically loaded models (.pkl).