Retraining updates a model’s learned parameters so it can reflect changing markets, customer behavior, regulations, or fraud patterns. It may finalize a credit model after validation or incorporate outcomes from fraud investigations. section 1.4.2 section 7.4.4
Production retraining depends on validation and ongoing oversight:
- For credit scoring, refit the model on the full WoE-coded training dataset after cross-validation establishes stable hyperparameters; preserve an isolated out-of-time hold-out set for regulatory validation. section 5.3.7
- For fraud detection, use investigation outcomes and fresh data to update models on a schedule. Monitor performance, drift, business impact, fairness, and compliance. section 7.4.4 section 8.1.3
Listing 5.18 Training the final model on all data
X_final = df_woe.drop(columns=["target"])
y_final = df_woe["target"]
from xgboost import XGBClassifier
final_model = XGBClassifier(
n_estimators=300,
max_depth=4,
learning_rate=0.2,
n_jobs=-1,
random_state=2023
)
final_model.fit(X_final, y_final)
print("Final model trained on the full WOE-coded dataset.")
The listing shows a final XGBoost classifier fit on the full WoE-coded dataset. section 5.3.7
