Context

LendWise built a binary classification model to predict whether a small-business loan applicant will default within 12 months. The model is currently used to support underwriting decisions, but leadership is concerned that the team reports only accuracy even though default cases are relatively rare.

Current Performance

Evaluation was run on a holdout set of 20,000 applications with a 10% default rate.

The Problem

The underwriting team wants to know which evaluation metrics should be prioritized for this classification problem and whether the current model is acceptable for deployment. Missing a true defaulter creates financial loss, while incorrectly flagging a safe applicant reduces approval volume and hurts customer experience.

Requirements

1. Explain which classification metrics you would use and why accuracy alone is insufficient.
2. Interpret the current metric values in the context of loan default prediction.
3. Discuss the tradeoff between precision and recall for this use case.
4. Recommend which metric(s) should be primary for model selection and threshold tuning.
5. Suggest concrete next steps to improve evaluation and model performance.

Constraints

• Each false negative default costs about $18,000 in expected loss.
• Each false positive rejection costs about $1,200 in lost contribution margin.
• Risk policy requires model outputs to be explainable enough for manual review.