projects / 2026 / Study
Credit Scorecard & Survival-Causal Study
An interpretable WOE/IV credit scorecard benchmarked against gradient boosting, plus survival and causal analysis of time-to-default on a public lending dataset.
What it is
A credit-risk study on a public lending dataset, in three parts.
The scorecard
An interpretable Logistic Regression scorecard with WOE/IV binning, benchmarked against a gradient-boosted model on AUC, KS and Gini. Class imbalance was handled with SMOTE and cost-sensitive thresholding, which meaningfully improved minority-class recall over the baseline.
Time to default
A Cox proportional-hazards model on time-to-default, to see which borrower and loan attributes change when a loan defaults, not only whether it does.
Does changing the loan term help?
A propensity-matched, difference-in-differences analysis estimating the effect of loan-term modifications on default risk.