Skip to content
BENIBIX Learn

Accuracy vs Precision

People often use accuracy and precision as if they mean the same thing. In machine learning they do not. Accuracy is the share of all answers that are right. Precision looks only at the yes answers, and asks how many of them were right. A model can be 99% accurate and still have low precision, when real yes cases are rare.

AI📏 Evaluation & Benchmarks

Accuracy (Machine Learning)

The share of a model's answers that are right. Nine right out of ten is 90%.

Accuracy is the share of a model's answers that are right, out of all its answers. If a model labels 100 photos and gets 90 right, its accuracy is 90%. It is the easiest score to understand, so it is often the first one people look at. But it can fool you when one answer is much more common than the others. Then a model can score high while missing nearly every case that matters.

Example
Lucy's model checks 1,000 X-rays and gets 970 right, so its accuracy is 97%. But only 30 X-rays showed the illness, and it missed 20 of them.
Why it matters
It is a quick first check that anyone can read. Just never let it be the only one.
Full entry
AI📏 Evaluation & Benchmarks

Precision (Machine Learning)

When a model says yes, how often it is right. Few false alarms means high precision.

Precision is how often a model is right when it says yes. Out of all the cases it marked as yes, it is the share that really were yes. A spam filter with 95% precision is wrong only 5 times in 100 when it calls something spam. High precision means few false alarms. It matters most when a false alarm is costly, such as sending a real email to the spam folder.

Example
Emma's spam filter flags 200 emails. 190 really are spam and 10 are real mail, so its precision is 95%.
Why it matters
It tells you how far you can trust a yes. That matters when acting on a wrong yes hurts someone.
Full entry

In one line

Accuracy (Machine Learning)
The share of a model's answers that are right. Nine right out of ten is 90%.
Precision (Machine Learning)
When a model says yes, how often it is right. Few false alarms means high precision.