03 — Model Training & Selection
Trains and compares regression and classification models for daily max demand prediction, with SHAP-based feature importance analysis.
xgboost regression classification shap model-selection
Pythonscikit-learnxgboostlightgbmshapmatplotlib
Key Findings
- ● XGBoost regression outperforms classification approaches by avoiding the extreme class imbalance problem
- ● Temperature features dominate SHAP importance, followed by demand momentum and calendar features
- ● Regression approach naturally provides uncertainty quantification via prediction intervals