Evaluate Metrics for Loan Default Model | Dataford Interview Questions

Context

LendWise uses a binary classification model to predict whether a personal loan applicant will default within 12 months. The model is used to support approval decisions, but risk leaders are concerned that the team is reporting only overall accuracy while default losses remain high.

Current Performance

Metric	Validation Set	Notes
Accuracy	0.91	High due to class imbalance
Precision	0.68	Of predicted defaults, 68% were correct
Recall	0.42	Model catches less than half of actual defaults
F1 Score	0.52	Weak balance between precision and recall
AUC-ROC	0.79	Moderate ranking performance
Default rate in data	0.10	10% positive class
Threshold	0.50	Current decision cutoff

The Problem

The model appears strong if judged only by accuracy, but it may be underperforming on the business objective: identifying risky applicants before loans are issued. The hiring manager wants to know whether the candidate understands which metrics matter, how to interpret them together, and what tradeoffs they imply.

Requirements

Explain which metrics should be used to evaluate this model and why accuracy alone is insufficient.
Interpret the current metric values in business terms.
Use the confusion matrix implications to describe the cost of false positives vs. false negatives.
Recommend which metric(s) LendWise should optimize for if the goal is reducing credit losses.
Suggest practical ways to improve model performance or evaluation.

Constraints

Each false negative leads to an average loss of $4,200.
Each false positive incorrectly rejects a profitable applicant worth $300 in expected margin.
Risk policy requires a transparent evaluation approach for model review.

Problem

Context

Current Performance

Metric	Validation Set	Notes
Accuracy	0.91	High due to class imbalance
Precision	0.68	Of predicted defaults, 68% were correct
Recall	0.42	Model catches less than half of actual defaults
F1 Score	0.52	Weak balance between precision and recall
AUC-ROC	0.79	Moderate ranking performance
Default rate in data	0.10	10% positive class
Threshold	0.50	Current decision cutoff

The Problem

Requirements

Explain which metrics should be used to evaluate this model and why accuracy alone is insufficient.
Interpret the current metric values in business terms.
Use the confusion matrix implications to describe the cost of false positives vs. false negatives.
Recommend which metric(s) LendWise should optimize for if the goal is reducing credit losses.
Suggest practical ways to improve model performance or evaluation.

Constraints

Each false negative leads to an average loss of $4,200.
Each false positive incorrectly rejects a profitable applicant worth $300 in expected margin.
Risk policy requires a transparent evaluation approach for model review.

Problem

Context

Current Performance

Metric	Validation Set	Notes
Accuracy	0.91	High due to class imbalance
Precision	0.68	Of predicted defaults, 68% were correct
Recall	0.42	Model catches less than half of actual defaults
F1 Score	0.52	Weak balance between precision and recall
AUC-ROC	0.79	Moderate ranking performance
Default rate in data	0.10	10% positive class
Threshold	0.50	Current decision cutoff

The Problem

Requirements

Explain which metrics should be used to evaluate this model and why accuracy alone is insufficient.
Interpret the current metric values in business terms.
Use the confusion matrix implications to describe the cost of false positives vs. false negatives.
Recommend which metric(s) LendWise should optimize for if the goal is reducing credit losses.
Suggest practical ways to improve model performance or evaluation.

Constraints

Each false negative leads to an average loss of $4,200.
Each false positive incorrectly rejects a profitable applicant worth $300 in expected margin.
Risk policy requires a transparent evaluation approach for model review.

Problem

Context

Current Performance

Metric	Validation Set	Notes
Accuracy	0.91	High due to class imbalance
Precision	0.68	Of predicted defaults, 68% were correct
Recall	0.42	Model catches less than half of actual defaults
F1 Score	0.52	Weak balance between precision and recall
AUC-ROC	0.79	Moderate ranking performance
Default rate in data	0.10	10% positive class
Threshold	0.50	Current decision cutoff

The Problem

Requirements

Explain which metrics should be used to evaluate this model and why accuracy alone is insufficient.
Interpret the current metric values in business terms.
Use the confusion matrix implications to describe the cost of false positives vs. false negatives.
Recommend which metric(s) LendWise should optimize for if the goal is reducing credit losses.
Suggest practical ways to improve model performance or evaluation.

Constraints

Each false negative leads to an average loss of $4,200.
Each false positive incorrectly rejects a profitable applicant worth $300 in expected margin.
Risk policy requires a transparent evaluation approach for model review.