Precision and Recall

In one line

Two complementary classification metrics — precision is “of what I flagged, how much was right” and recall is “of what was actually there, how much did I catch”.

What it actually means

For a binary classifier with predictions and ground truth, you can sort outcomes into true positives, false positives, true negatives, and false negatives. Precision is TP / (TP + FP). Recall is TP / (TP + FN). They trade off against each other: lowering the decision threshold catches more positives (higher recall) but flags more junk (lower precision). The right balance depends on the cost of each error type — false positives in spam filtering are mildly annoying, false negatives in cancer screening are catastrophic.

Why it matters

Accuracy is misleading on imbalanced data. A 99% accurate fraud detector that flags nothing is “accurate” and useless. Precision and recall tell you what’s actually happening, and they’re how product, legal, and risk teams will phrase requirements: “we need 90% recall at 70% precision”. Get fluent with them.

Example

TP = 70, FP = 30, FN = 20, TN = 880
precision = 70 / (70 + 30) = 0.70
recall    = 70 / (70 + 20) = 0.78

You’ll hear it when

• Tuning a classifier threshold for a business requirement.
• Reviewing model cards and confusion matrices.
• Building a content moderation, fraud, or anomaly detection system.
• Evaluating a retriever in a RAG pipeline (recall@k).
• Negotiating SLAs with downstream stakeholders.