# Data Interview Must-Know: Precision vs Recall vs F1 (Stop Mixing Them Up)

<img src="https://bugfree-s3.s3.amazonaws.com/mermaid_diagrams/image_1776100811056.png" alt="Precision vs Recall vs F1" width="700" style="max-width:100%;height:auto;">

## Precision vs Recall vs F1 — Quick, interview-ready explanations

When interviewers ask about evaluation metrics, they want clear definitions, when to use each, and a short justification tied to business cost. Here are compact explanations you can say confidently.

### Confusion matrix reminder

|                | Predicted Positive | Predicted Negative |
|----------------|-------------------:|-------------------:|
| Actual Positive | TP                | FN                |
| Actual Negative | FP                | TN                |

- TP = true positives, FP = false positives, FN = false negatives, TN = true negatives.

### Precision
- Formula: `Precision = TP / (TP + FP)`
- Meaning: Of the examples predicted positive, what fraction were actually positive?
- Use when false positives are costly: e.g., marking good email as spam, recommending irrelevant products, or flagging innocent users as fraudulent.
- Interview line: “Optimize precision when you must avoid false alarms.”

### Recall (a.k.a. Sensitivity)
- Formula: `Recall = TP / (TP + FN)`
- Meaning: Of all actual positives, how many did the model catch?
- Use when false negatives are costly: e.g., missing a disease, failing to detect a defect, or not flagging a dangerous event.
- Interview line: “Optimize recall when missing positives has a high cost.”

### F1 score
- Formula: `F1 = 2 * (Precision * Recall) / (Precision + Recall)`
- Meaning: The harmonic mean of precision and recall — balances both.
- Use when you need a single metric and care about both precision and recall (common with imbalanced classes like fraud detection).
- Interview line: “Use F1 for a balanced view when both error types matter and classes are imbalanced.”

Tip: if you care more about recall than precision, use F-beta (beta > 1) to weight recall higher; if precision matters more, use beta < 1.

### Trade-offs & practical advice
- Increasing model threshold → usually increases precision, decreases recall. Lowering threshold → increases recall, decreases precision.
- Always align the metric with business costs (false positive vs false negative consequences). Explain the cost trade-off in the interview.
- For highly imbalanced datasets, prefer Precision-Recall curves over ROC curves — PR curves are more informative when positives are rare.
- When reporting, show the confusion matrix, chosen operating point (threshold), and why that metric matches the problem.

### Short, interview-ready answers
- "Precision measures how many predicted positives are correct; use it when false positives are expensive."
- "Recall measures how many actual positives we caught; use it when false negatives are dangerous."
- "F1 balances both; use it for imbalanced data or when you want a single summary metric."

Use these lines, back them with a short example from your past work (or a concise hypothetical) and you’ll sound precise, practical, and interview-ready.