Predicting JOSAA College Admissions Using an Ensemble of Year-Trend and Round-Progression Models From Automated Data Collection to Multi-Round Forecasting

Key Points

• The aim is to predict JoSAA and CSAB college admission closing ranks using automated data collection and forecasting models.
• Utilized Playwright-based scrapers for historical data collection from JoSAA and CSAB archives (2016-2025).
• Developed a per-slot GP-MLP ensemble model for predicting closing ranks across multiple rounds (R1-R6).
• Implemented a Streamlit web application to provide predictions for students.
• GP-MLP ensemble model achieved an average MAE of 2,608 on JoSAA predictions (2022-2025), outperforming other models.
• CSAB ensemble produced an MAE of 42,318 on 2025 predictions, indicating a 15.1% improvement over GP RBF.
• JoSAA closing ranks exhibited mean-reversion, with historical medians outperforming linear extrapolation by 23% on average.