Accuracy hides failure. A fraud detector that flags nothing is right almost every time, yet useless. Precision and recall expose the tradeoff. Precision asks how many predicted positives are actually positive. Recall asks how many actual positives were found.

The two pull against each other. Raising the threshold catches fewer false positives but misses more true positives, improving precision and hurting recall. Lowering it does the reverse. The right balance depends on which error costs more.

Choosing between them

• High precision, when false positives are expensive
• High recall, when missing positives is dangerous
• F1 score, when both matter equally
• Precision-recall curve, for threshold selection

Cancer screening favours recall. Spam filtering favours precision. Neither metric alone tells the whole story, which is why both appear together in serious evaluation reports.