Business Context

LendWise, a mid-size digital lending platform, wants to use machine learning in its credit risk analysis workflow. The goal is to predict whether a newly approved loan applicant will default within 90 days so the risk team can prioritize manual review and adjust approval thresholds.

Dataset

You are given a historical training dataset of past loan applications and repayment outcomes.

• Size: 120K loan applications, 32 input features
• Target: default_90d — whether the borrower defaults within 90 days of origination
• Class balance: 14% positive class
• Missing data: ~8% missing in income-related fields, ~5% missing in bureau variables

Success Criteria

A good solution should achieve strong ranking quality and usable recall for the risk team. Target performance is ROC-AUC >= 0.82 and F1 >= 0.55 at an operational threshold, while keeping the model explainable enough for analysts to review decisions.

Constraints

• Predictions must run in batch daily and score up to 200K applications in under 10 minutes.
• The risk team needs interpretable drivers of default risk.
• Training should use standard Python ML tooling and be maintainable by analysts.

Deliverables

1. Build a supervised classification pipeline for default_90d.
2. Explain how you would preprocess mixed data types and missing values.
3. Compare at least one interpretable baseline with one stronger non-linear model.
4. Evaluate the model using appropriate metrics for imbalanced classification.
5. Recommend a thresholding strategy for business use and describe how you would deploy and monitor the model.