Why can an overconfident weak classifier distort a soft-voting ensemble more than its hard vote suggests?
Classic MLLevel: IntermediateProbability ensemble reliability
Why can an overconfident weak classifier distort a soft-voting ensemble more than its hard vote suggests?
Hints
- A probability contributes magnitude, not just a label.
- Poor calibration can dominate the aggregate.
Answer
Soft voting averages probabilities, so extreme confidence affects the mean even when the class vote is unchanged. Validate calibration and weights rather than averaging incompatible confidence scales blindly.
Source and licence: scikit-learn 1.7.2 — ensemble · BSD-3-Clause · answer: Quantoros editorial