Should every binary classifier tune its threshold to maximize balanced accuracy?
Classic MLLevel: IntermediateDecision utility
Should every binary classifier tune its threshold to maximize balanced accuracy?
Hints
- A default is not a requirement.
- Error costs can be asymmetric.
Answer
No. It is a default objective, not a universal business utility. Choose a metric or explicit cost function reflecting false positives, false negatives and operational constraints.
Source and licence: scikit-learn 1.7.2 — classification threshold · BSD-3-Clause · answer: Quantoros editorial