Decision Threshold

noun

say it: “dih-SIZH-un THRESH-hohld” /dɪˈsɪʒən ˈθrɛʃˌhoʊld/

The cut-off value above which a classifier's estimated probability counts as a 'yes'.

What is a decision threshold?

The decision threshold is the probability at which a classifier switches its answer from no to yes; predict() uses 0.5. Lowering it catches more of the positive class but raises more false alarms; raising it does the opposite. You set your own by comparing predict_proba() with it.

chances = [0.15, 0.35, 0.55, 0.92]    # predict_proba()'s fraud column
print([p >= 0.5 for p in chances])   # [False, False, True, True]
print([p >= 0.3 for p in chances])   # [False, True, True, True]: one more flagged

In a sentence

  • “predict() uses a decision threshold of 0.5.”
  • “Dropping the decision threshold to 0.3 caught two more frauds and five more false alarms.”

Level: Journeyman · More AI Basics terms at this level

Updated