Business Context

LendWise, a consumer lending fintech processing roughly 120K loan applications per month, wants a machine learning model to predict whether an applicant will default within 12 months of loan origination. The model will support underwriting decisions and risk-based pricing, so it must improve risk separation without creating an overly complex production system.

Dataset

You are given a historical dataset of funded loans from the last 3 years.

• Rows: 420K funded loans, 40 features
• Target: default_12m — whether the customer becomes 90+ days past due or charged off within 12 months
• Class balance: 11.6% default, 88.4% non-default
• Missing data: 18% missing in employment_length, 9% in annual_income, and light missingness (<3%) in several bureau fields

Success Criteria

A good solution should achieve meaningful lift over a logistic regression baseline, with strong ranking quality for underwriting. Target performance is ROC-AUC >= 0.82, PR-AUC >= 0.42, and a top-decile lift > 2.5 on a held-out test set.

Constraints

• Predictions must be generated in <50 ms per application in an online API.
• Risk and compliance teams require feature-level explanations.
• The model should be retrained monthly and remain stable under moderate feature drift.

Deliverables

1. Build a binary classification pipeline to predict 12-month default.
2. Explain model choice, preprocessing, and feature engineering decisions.
3. Define a validation strategy that avoids leakage from future-originated loans.
4. Report evaluation metrics and recommend an operating threshold for underwriting.
5. Provide production-ready Python code for training and inference.