Predict Loan Default with Gradient Boosting

Business Context

LendPro, a digital consumer lending platform processing roughly 300K applications per quarter, wants to reduce default losses without materially lowering approval volume. Build a model that predicts whether a newly issued loan will default within 12 months so the risk team can improve underwriting decisions.

Dataset

You are given a historical loan-origination dataset with one row per funded loan.

• Historical window: 24 months of funded loans
• Observation unit: funded loan at origination time
• Target: whether the loan defaults within 12 months of funding

Success Criteria

A strong solution should outperform a logistic regression baseline and achieve enough ranking quality for underwriting. Target ROC-AUC >= 0.82, PR-AUC >= 0.38, and recall >= 0.70 at precision >= 0.45 on the held-out test set.

Constraints

• Predictions must be generated in <100 ms per application in an online API.
• The risk team needs feature-level explanations for adverse action review.
• Training budget is moderate; a single daily retraining job is acceptable.
• Missing values and class imbalance must be handled explicitly.

Deliverables

1. Select and justify an algorithm for this default prediction problem.
2. Build a full preprocessing and training pipeline.
3. Explain how you would handle imbalance, missing data, and categorical variables.
4. Evaluate the model with business-relevant metrics and threshold selection.
5. Provide an interpretation approach suitable for underwriting and model monitoring in production.