Context

LendWise built a binary classification model to predict whether a small-business loan applicant will default within 12 months. The model is used to support approval decisions, but the credit team is concerned that the current evaluation focuses too heavily on overall accuracy while missing risky applicants.

The model was trained on 200,000 historical applications and evaluated on a holdout set of 50,000 recent applications. In this test set, 5,000 applicants actually defaulted and 45,000 did not.

Current Performance

The Problem

Leadership sees 91% accuracy and believes the model is performing well enough for rollout. However, the risk team points out that the model is still missing a large share of actual defaulters. You need to determine which metrics matter most, how to interpret the current results, and what changes should be recommended before deployment.

Requirements

1. Explain which evaluation metrics you would prioritize for this model and why.
2. Interpret what the current metric values imply about model behavior.
3. Use the confusion matrix to describe the business impact of false positives vs. false negatives.
4. Recommend specific steps to improve evaluation and model performance.
5. State whether accuracy alone is sufficient for launch approval.

Constraints

• A false negative can lead to significant credit losses.
• A false positive rejects a potentially good customer and reduces revenue.
• The underwriting team prefers an interpretable decision threshold and cannot manually review more than 3,500 flagged applications per month.