Business Context

LendWise, a digital consumer lending platform processing ~120K loan applications per month, wants a supervised learning model to predict whether an applicant will default within 12 months. The risk team needs a model that improves approval decisions while remaining explainable enough for audit and policy review.

Dataset

You are given a historical labeled dataset where each row represents one funded loan and the target indicates whether the borrower defaulted. This is a standard supervised learning problem because the model is trained on input features and known outcomes.

• Size: 96K funded loans, 30 features
• Target: Binary label — default within 12 months (1) vs no default (0)
• Class balance: Moderately imbalanced, 18% default / 82% non-default
• Missing data: ~9% missing in income-related fields, ~6% missing in bureau freshness variables, <2% elsewhere

Success Criteria

A good solution should achieve ROC-AUC >= 0.82, F1 >= 0.58, and recall >= 0.70 for the default class at an operational threshold chosen with business tradeoffs in mind.

Constraints

• Predictions must run in <50 ms per application in an online approval flow.
• The model must support feature importance or reason codes for risk analysts.
• Retraining should be feasible on a monthly cadence with standard Python tooling.

Deliverables

1. Define supervised learning in the context of this dataset and explain why this is a supervised classification task.
2. Build a baseline logistic regression model and a stronger tree-based model.
3. Create a preprocessing pipeline for missing values, categorical variables, and numeric scaling where appropriate.
4. Evaluate models using cross-validation and a held-out test set.
5. Recommend a final model, decision threshold, and monitoring plan for production use.