Implement a modeling pipeline evaluation tool

I am doing a project for credit risk using Python. I'd love a sanity check on my pipeline and some opinions on gaps or mistakes or anything which might improve my current modeling pipeline. Also would be grateful if you can score my current pipeline out of 100% as per your assessment :) **My current pipeline** 1. Import data 2. Missing value analysis — bucketed by % missing (0–10%, 10–20%, …, 90–100%) 3. Zero-variance feature removal 4. Sentinel value handling (-1 to NaN for categoricals) 5. Leakage variable removal (business logic) 6. Target variable construction 7. create new features 8. Correlation analysis (numeric + categorical) drop one from each correlated pair 9. Feature-target correlation check — drop leaky features or target proxy features 10. Train / test / out-of-time (OOT) split 11. WoE encoding for logistic regression 12. VIF on WoE features — drop features with VIF > 5 13. Drop any remaining leakage + protected variables (e.g. Gender) 14. Train logistic regression with cross-validation 15. Train XGBoost on raw features 16. Evaluation: AUC, Gini, feature importance, top feature distributions vs target, SHAP values 17. Hyperparameter tuning with Optuna 18. Compare XGBoost baseline vs tuned 19. Export models for deployment Improvements I'm already planning to add * Outlier analysis * Deeper EDA on features * Missingness pattern analysis: MCAR / MAR / MNAR * KS statistic to measure score separation * PSI (Population Stability Index) between training and OOT sample to check for representativeness of features