Business Context

NorthStar Lending is a mid-sized digital lender that processes roughly 250,000 personal loan applications per year. The risk team wants a machine learning model to predict whether an approved applicant will default within 12 months so underwriting rules can be improved without materially slowing application review.

Dataset

You are given a historical loan dataset built from approved applications and repayment outcomes.

• Size: 120,000 loan records, 38 input features
• Target: default_12m — whether the borrower defaulted within 12 months of origination
• Class balance: 14% positive class, 86% negative class
• Missing data: 8% missing in employment and housing fields, 3% missing in bureau attributes, and some missingness is informative

Success Criteria

A good solution should outperform a simple logistic regression baseline, achieve strong ranking quality for risk-based decisioning, and provide enough interpretability for the credit policy team to review the main drivers of risk.

Constraints

• Batch scoring only; inference should complete in under 10 minutes for 50,000 applications
• Model outputs must be explainable at a feature level
• Training should use only data available at application time
• Avoid leakage from post-origination repayment behavior

Deliverables

1. Build a binary classification pipeline to predict 12-month default.
2. Compare at least one baseline model and one stronger non-linear model.
3. Describe preprocessing, feature engineering, and validation strategy.
4. Report evaluation metrics and choose a decision threshold for underwriting.
5. Summarize key tradeoffs between model quality, interpretability, and operational use.