CWC Flood Forecasting System

XGBoost and LSTM models for 20-hour water level forecasting, developed for the Central Water Commission.

Key Highlights

• XGBoost model achieves 88.15% accuracy (within ±0.15m) — recommended for operational use
• Rich feature engineering: lags, rolling stats, rainfall aggregates, temporal encodings, deltas
• Benchmarked against 3 LSTM architectures (Seq-to-Vec, Multi-Step, Seq-to-Seq)