Precision

noun

say it: “prih-SIZH-un” /prɪˈsɪʒən/

Of all the items a model labeled as positive, the share that really were positive.

What is precision for a classifier?

Precision is, of the examples a model flagged, the share that really were the positive class: true positives divided by everything it flagged, TP / (TP + FP). High precision means few false alarms. It says nothing about the positives the model missed; recall does.

from sklearn.metrics import precision_score
actual = [1, 0, 0, 1, 1, 0]    # 1 = fraud
guessed = [1, 0, 1, 0, 1, 0]
print(round(precision_score(actual, guessed), 3))   # 0.667: 2 of 3 flags were fraud

In a sentence

  • “Raising the threshold improved precision but let more fraud through.”
  • “A precision of 0.667 meant two flags in three were real fraud.”

Level: Journeyman · More AI Basics terms at this level

Updated