04 — ML-Based RDII Detection

Trains XGBoost and LSTM models for RDII prediction, performs autoencoder-based anomaly detection for I&I period classification, and validates against conductivity-based dilution analysis.

xgboost lstm anomaly-detection shap conductivity

Pythonpandasnumpyscikit-learnxgboosttensorflowshapmatplotlib

Key Findings

● XGBoost and LSTM both achieve strong RDII prediction, with LSTM capturing longer temporal dependencies
● SHAP analysis confirms rainfall lag and snowmelt as dominant features for inflow plants, antecedent moisture for infiltration plants
● Multi-method consistency check across CCF, RDII decomposition, ML prediction, and conductivity validates detection reliability

Notebook

Open in Google Colab Download Notebook (.ipynb)