SEP Occurrence Forecasting with XGBoost

Three notebooks use gradient-boosted trees on the same split as the neural networks.

• XGBoost classifier: training with early stopping, verification scores at three thresholds, and bootstrap intervals.

• Repeated validation: how much the scores vary across 15 cross-validation folds.

• SHAP attributions: which predictors the model relies on, and whether that ranking survives a change of random seed.