Context

LendWise built a binary classification model to predict whether a small-business loan application will default within 12 months. The team initially reported only accuracy, but risk leaders are concerned that this may hide poor performance on the minority default class.

Current Performance

The model was evaluated on 20,000 recent applications with a 10% default rate.

The Problem

Although the model appears strong on accuracy, it misses a large share of actual defaults. The credit policy team wants to know which metrics should drive evaluation and whether the current model is acceptable for production decisioning.

Requirements

1. Explain which classification metrics are most important in this setting and why.
2. Interpret the current metric values together rather than in isolation.
3. Identify what the confusion matrix says about business risk.
4. Recommend which additional validation or threshold analysis should be done before deployment.
5. Propose specific ways to improve the model or decision policy.

Constraints

• Approving a defaulter creates an average loss of $18,000.
• Incorrectly rejecting a good applicant creates an average opportunity cost of $1,200.
• Regulators require explainable decision policies and stable monitoring after launch.