Context

LendWise uses a binary classification model to predict whether a personal loan applicant will default within 12 months. The team reports strong overall accuracy, but the credit risk manager is concerned that too many risky borrowers are still being approved.

Current Performance

Confusion matrix on 10,000 validation applicants:

The Problem

The model looks good if you focus on accuracy alone, but missed defaults are expensive because they lead to loan losses. The team needs a better evaluation of whether this model is actually suitable for deployment and what should be improved.

Requirements

1. Interpret the model’s performance using the provided metrics, not accuracy alone.
2. Explain what the confusion matrix says about business risk.
3. Identify the main weakness of the current model and why it matters.
4. Recommend which additional validation or threshold analysis you would run before launch.
5. Propose practical changes to improve model performance for this use case.

Constraints

• Each false negative default costs LendWise about $4,000 in expected loss.
• Each false positive rejection costs about $300 in lost profit and customer dissatisfaction.
• The underwriting team prefers a simple thresholding policy that can be audited easily.