Context

MediScan built a binary classifier to flag chest X-rays for possible pneumonia so radiologists can prioritize urgent cases. After deployment, the hospital noticed that the model catches many true pneumonia cases, but it also sends a large number of healthy scans for review.

Current Performance

The Problem

Clinical leadership wants to understand whether this model is appropriately tuned. Missing a pneumonia case is costly, but too many false alarms increase radiologist workload and delay other urgent reads. The team needs a clear explanation of precision and recall, what these values imply here, and whether the threshold should change.

Requirements

1. Define precision and recall using this model's outputs.
2. Interpret what the current metric values mean in business and clinical terms.
3. Explain why accuracy alone is not sufficient in this setting.
4. Use the confusion matrix to describe the tradeoff between false positives and false negatives.
5. Recommend whether MediScan should prioritize higher precision, higher recall, or keep the current balance.

Constraints

• Radiology team can absorb at most 320 flagged scans per day.
• A missed pneumonia case has much higher cost than an unnecessary review.
• Any threshold change must preserve recall above 0.85.