What does the classification threshold control, and who should set it?
- A The probability cut-off for a positive prediction, determined by the relative cost of each error type
- B The model architecture
- C The learning rate
- D The number of features
Answer
The probability cut-off for a positive prediction, determined by the relative cost of each error type
The default of 0.5 is arbitrary. Presenting the precision-recall trade-off lets the business choose the operating point.





