Context

LendWise built a binary classification model to predict whether a personal loan applicant will default within 12 months. The team wants to deploy the model to support underwriting decisions, but risk leadership wants a clear validation plan before launch because false approvals increase credit losses while false declines reduce revenue and create customer friction.

Current Performance

The model was trained on 180,000 historical applications and evaluated on a holdout set of 20,000 recent applications.

The Problem

The model looks directionally useful, but performance drops from training to test, and recall is materially lower than precision. The lending team needs to know whether the model is reliable enough for deployment, what additional validation is required, and whether the current threshold is appropriate for business use.

Requirements

1. Explain what these metrics suggest about generalization and model quality.
2. Describe the validation steps you would complete before deployment.
3. Identify what additional analyses you would run beyond aggregate metrics.
4. Recommend whether you would deploy now, deploy with safeguards, or delay deployment.
5. Propose specific improvements to increase confidence in the model.

Constraints

• False negatives (missed defaulters) cost about $4,200 per loan.
• False positives (incorrectly declining good borrowers) cost about $350 in expected contribution margin.
• The model must be explainable enough for risk and compliance review.