Final model: XGBoost (F1-optimised via focused tuning)

Best params: {'model__colsample_bytree': 0.8, 'model__learning_rate': 0.1, 'model__max_depth': 6, 'model__n_estimators': 400, 'model__scale_pos_weight': 4.182552504038772, 'model__subsample': 1.0}

Test F1: 0.8041 | Precision: 0.8440 | Recall: 0.7677 | ROC-AUC: 0.9580