Fraud is encoded as 0 rather than 1. What can go wrong when tuning a threshold with the default F1 scorer?
Classic MLLevel: IntermediatePositive label configuration
Fraud is encoded as 0 rather than 1. What can go wrong when tuning a threshold with the default F1 scorer?
Hints
- A numeric code is not business meaning.
- The scorer carries its own label convention.
Answer
The scorer may optimize the wrong positive class. Build a scorer with the intended pos_label and confirm that decision scores and probabilities are interpreted consistently.
Source and licence: scikit-learn 1.7.2 — classification threshold · BSD-3-Clause · answer: Quantoros editorial