Knowledge Maxxing← Back to topic library
ML MaxxingIntermediate75 sec
Classification threshold
A classification threshold is the probability cutoff above which a model's predicted score is converted into a positive class prediction.
Prerequisite: Precision
What matters
Moving the threshold trades off precision and recall without retraining the model at all, since the same underlying scores get sorted into positive or negative differently.
Example
Lowering a fraud model's threshold from 0.9 to 0.5 flags more transactions as fraud, catching more real cases but likely triggering more false alarms too.
Common misconception
A classification threshold of 0.5 is not always the right default; the best threshold depends on the relative cost of false positives versus false negatives for that task.
Recall question
What happens to precision and recall when a classification threshold is lowered?
Keywords and related concepts
classification thresholddecision boundaryprecision-recall trade-offroc curve