Context

LendWise uses a binary classification model to predict whether a personal loan applicant will default within 12 months. The model was recently deployed to support underwriting decisions, but the credit risk team is concerned that headline accuracy looks strong while business outcomes remain mixed.

Current Performance

Additional validation-set counts (20,000 applicants): TP = 912, FP = 558, FN = 1,488, TN = 17,042.

The Problem

The model appears better on accuracy and precision, but it is missing many actual defaulters. The underwriting team wants to know whether the model is truly effective and whether the current decision threshold is appropriate.

Requirements

1. Assess the model using the metrics above rather than relying on accuracy alone.
2. Explain what the confusion matrix says about business risk.
3. Identify the main tradeoff between precision and recall in this setting.
4. Recommend how you would validate whether the threshold should be changed.
5. Propose specific next steps to improve model performance and evaluation.

Constraints

• Approving a defaulter has a much higher cost than rejecting a good applicant.
• The model is used as a decision-support tool, so false positives also create revenue loss.
• The team can retrain monthly, but policy changes require clear metric justification.