Research case study
Mortgage Default Prediction
Default prediction modeling for mortgage risk.
Focus
Python · XGBoost
Methods
SHAP · Risk
01 / Context
The research question
How accurately can we predict mortgage defaults using loan-level characteristics, and which features drive risk?
02 / Method
Research design
Built a classification pipeline with Logistic Regression, Random Forest, and XGBoost on loan-level data with feature engineering for LTV, DTI, and credit history.
03 / Evaluation
How it was tested
- Loan-level data, temporal train/test split by vintage
- Metrics: AUC, precision-recall, SHAP feature importance
04 / Findings
What the work showed
In the reported temporal holdout, XGBoost ranked strongest among the tested models and SHAP highlighted LTV and credit score as leading drivers.
05 / Next iteration
Where I would take it next
Incorporate macroeconomic covariates (unemployment, HPI) and test on out-of-sample vintage cohorts.