Precision
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The share of a model's positive predictions that were actually correct – the metric to watch when a false alarm costs more than a missed case.
What is Precision?
Precision is the share of a classification model’s positive predictions that were actually correct: true positives divided by everything the model flagged as positive. A precision of 0.9 means nine out of ten alarms were real.
In machine learning the term has nothing to do with everyday carefulness or with numeric precision (the bit-width of numbers, as in quantization); it measures one thing only, how trustworthy a model’s positive calls are.
How Precision Works
Every positive prediction is either a true positive (TP), a correct flag, or a false positive (FP), a false alarm. Precision is the fraction of flags that were right:
\[ \text{Precision} = \frac{TP}{TP + FP} \]Both counts come straight from the confusion matrix. Note what precision ignores: positives the model failed to flag at all. A model that flags a single case, correctly, scores a perfect 1.0 while missing everything else.
Precision vs Recall
Precision asks how many of the model’s flags were right; recall asks how many of the real positives the model found. The two pull against each other: flag less and precision rises while recall falls, flag more and the reverse.
| Precision | Recall | |
|---|---|---|
| Question answered | Of everything flagged, how much was right? | Of all real positives, how many were found? |
| Error it penalizes | False positives (false alarms) | False negatives (misses) |
| Denominator | All predicted positives (TP + FP) | All actual positives (TP + FN) |
| Matters most when | A false alarm is expensive | A miss is expensive |
In cancer screening, a false negative means an untreated tumor, so recall leads. In spam filtering, a false positive buries a real message, so precision leads. When neither error is clearly cheaper, the F1 score combines the two.
Example of Precision
A hospital screens 1,000 patients; 50 actually have the disease. The model flags 80 patients: 40 truly sick (TP) and 40 healthy (FP).
Precision is 40 / 80 = 0.5. Half the flagged patients face follow-up tests for nothing, even though the model caught 40 of the 50 real cases (a recall of 0.8). Whether 0.5 is acceptable depends on the price of a false alarm, which is exactly the question precision is built to answer.
Related AI terms: Recall · F1 Score · Confusion Matrix · Model Evaluation · ROC Curve
Did you like the Precision gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About