Wellness Tourism Package - purchase prediction

Best model: XGBoost (selected by 5-fold CV F1). Decision threshold: 0.499 (tuned on out-of-fold training predictions).

Split	Accuracy	Precision	Recall	F1	ROC-AUC
Train	0.999	0.995	1.000	0.998	1.000
Test	0.922	0.829	0.748	0.786	0.941

Load it with joblib.load(...) -> dict with keys model, threshold, feature_columns, model_name.