MediScan uses a binary classifier to flag chest X-rays for possible pneumonia so radiologists can prioritize urgent cases. The team is concerned that the current model appears highly accurate when it predicts a positive case, but it may be missing too many actual pneumonia patients.
| Metric | Value |
|---|---|
| Precision | 0.91 |
| Recall | 0.42 |
| F1 Score | 0.57 |
| Accuracy | 0.95 |
| AUC-ROC | 0.84 |
| Threshold | 0.80 |
| Positive prevalence | 8% |
On a recent validation set of 10,000 X-rays, the model produced the following counts:
| Outcome | Count |
|---|---|
| True Positives | 336 |
| False Positives | 34 |
| False Negatives | 464 |
| True Negatives | 9,166 |
The radiology lead wants to understand what it means, in this context, for the classifier to have high precision but low recall. The team must decide whether the model is too conservative and whether threshold or workflow changes are needed.