Context

LendWise uses a binary classification model to predict whether a small-business loan applicant will default within 12 months. The risk team currently evaluates the model mainly with accuracy, but business leaders are debating whether precision, recall, F1-score, or ROC-AUC should drive model decisions because false approvals and false rejections have very different costs.

Current Performance

The model was evaluated on 20,000 recent applications, with a default rate of 8% (1,600 actual defaults).

The Problem

The Chief Risk Officer wants fewer bad loans approved, while the Growth team is concerned that too many good applicants are being declined. You need to explain which metric should be prioritized for different business goals and whether the current threshold is appropriate.

Requirements

1. Interpret what each metric says about this model in the context of loan approvals.
2. Explain when precision, recall, F1-score, and ROC-AUC should be used for this business problem.
3. Identify which metric is most appropriate for a risk-focused strategy versus a growth-focused strategy.
4. Recommend whether LendWise should keep or change the current decision threshold.
5. Describe what additional analysis you would run before changing the model or threshold.

Constraints

• A false negative (missed default) costs about $18,000 in expected losses.
• A false positive (declining a good applicant) costs about $1,200 in lost contribution margin.
• Regulators require the team to justify approval rules clearly and monitor model performance monthly.