Context

LendWise uses a binary classification model to predict whether a personal loan applicant will default within 90 days. The current model was recently deployed to support underwriting decisions, but business stakeholders are concerned that headline accuracy looks strong while default losses remain higher than expected.

Current Performance

The Problem

The underwriting team sees fewer applicants flagged as high-risk, but post-loan analysis shows too many actual defaulters are still being approved. You need to assess what to look at when evaluating this model beyond a single metric and explain whether the current model is actually better for the business.

Requirements

1. Interpret the current metrics together rather than in isolation.
2. Explain why high accuracy may be misleading in this setting.
3. Use the confusion matrix implications to discuss business tradeoffs between false positives and false negatives.
4. Recommend which additional analyses or validation checks you would perform before approving this model.
5. Propose specific actions to improve model performance or decision thresholding.

Constraints

• False negatives are costly because missed defaulters create direct credit losses.
• False positives reduce loan approvals and revenue from good borrowers.
• Risk policy requires a simple threshold-based decision rule for deployment.