Business Context

LendWise, a digital consumer lending platform processing ~250K applications per quarter, wants a model to predict whether an approved borrower will default within 90 days of origination. The goal is to improve underwriting decisions while keeping the model explainable for risk and compliance teams.

Dataset

• Rows: 320K historical loans, 40 features
• Target: default_90d — 1 if borrower misses payments and is charged off within 90 days, else 0
• Class balance: 11.6% positive, 88.4% negative
• Missing data: 18% missing in bank transaction features, 6% missing in bureau variables for thin-file applicants

Success Criteria

A solution is considered good enough if it achieves ROC-AUC >= 0.82, PR-AUC >= 0.42, and recall >= 0.70 at precision >= 0.35 on a held-out test set. The candidate should also show a clear, repeatable model-building process from raw data to threshold selection.

Constraints

• Predictions must be returned in <50 ms for online underwriting.
• The model must support reason codes / feature importance for adverse action review.
• Retraining is expected monthly using the latest approved-loan data.
• Avoid leakage from post-origination variables.

Deliverables

1. Define a general model-building process for this classification task.
2. Build and compare at least one baseline and one stronger model.
3. Explain preprocessing, feature engineering, validation, and leakage checks.
4. Select evaluation metrics and a decision threshold aligned to business risk.
5. Describe how the model would be deployed and monitored in production.