MediScan uses a binary classification model to flag patients as high risk for early-stage lung disease based on imaging and clinical features. The clinical team is concerned that the model looks strong on overall accuracy, but too many true cases may be going undetected.
| Metric | Current Model | Previous Model | Change |
|---|---|---|---|
| Accuracy | 0.94 | 0.91 | +0.03 |
| Precision | 0.78 | 0.70 | +0.08 |
| Recall | 0.61 | 0.74 | -0.13 |
| F1 Score | 0.68 | 0.72 | -0.04 |
| AUC-ROC | 0.87 | 0.85 | +0.02 |
| Positive prevalence | 8.0% | 8.0% | 0.00 |
The new model improved accuracy and precision, but recall fell from 74% to 61%. In this setting, missing a true positive can delay diagnosis and treatment. The hiring manager wants to assess whether you understand what recall means, how to interpret it alongside other metrics, and when it should be prioritized over accuracy or precision.